xref: /linux/net/ipv4/ip_gre.c (revision b04df400c30235fa347313c9e2a0695549bd2c8e)
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 void 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 	switch (type) {
150 	default:
151 	case ICMP_PARAMETERPROB:
152 		return;
153 
154 	case ICMP_DEST_UNREACH:
155 		switch (code) {
156 		case ICMP_SR_FAILED:
157 		case ICMP_PORT_UNREACH:
158 			/* Impossible event. */
159 			return;
160 		default:
161 			/* All others are translated to HOST_UNREACH.
162 			   rfc2003 contains "deep thoughts" about NET_UNREACH,
163 			   I believe they are just ether pollution. --ANK
164 			 */
165 			break;
166 		}
167 		break;
168 
169 	case ICMP_TIME_EXCEEDED:
170 		if (code != ICMP_EXC_TTL)
171 			return;
172 		data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */
173 		break;
174 
175 	case ICMP_REDIRECT:
176 		break;
177 	}
178 
179 	if (tpi->proto == htons(ETH_P_TEB))
180 		itn = net_generic(net, gre_tap_net_id);
181 	else
182 		itn = net_generic(net, ipgre_net_id);
183 
184 	iph = (const struct iphdr *)(icmp_hdr(skb) + 1);
185 	t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
186 			     iph->daddr, iph->saddr, tpi->key);
187 
188 	if (!t)
189 		return;
190 
191 #if IS_ENABLED(CONFIG_IPV6)
192        if (tpi->proto == htons(ETH_P_IPV6) &&
193            !ip6_err_gen_icmpv6_unreach(skb, iph->ihl * 4 + tpi->hdr_len,
194 				       type, data_len))
195                return;
196 #endif
197 
198 	if (t->parms.iph.daddr == 0 ||
199 	    ipv4_is_multicast(t->parms.iph.daddr))
200 		return;
201 
202 	if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
203 		return;
204 
205 	if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
206 		t->err_count++;
207 	else
208 		t->err_count = 1;
209 	t->err_time = jiffies;
210 }
211 
212 static void gre_err(struct sk_buff *skb, u32 info)
213 {
214 	/* All the routers (except for Linux) return only
215 	 * 8 bytes of packet payload. It means, that precise relaying of
216 	 * ICMP in the real Internet is absolutely infeasible.
217 	 *
218 	 * Moreover, Cisco "wise men" put GRE key to the third word
219 	 * in GRE header. It makes impossible maintaining even soft
220 	 * state for keyed
221 	 * GRE tunnels with enabled checksum. Tell them "thank you".
222 	 *
223 	 * Well, I wonder, rfc1812 was written by Cisco employee,
224 	 * what the hell these idiots break standards established
225 	 * by themselves???
226 	 */
227 
228 	const struct iphdr *iph = (struct iphdr *)skb->data;
229 	const int type = icmp_hdr(skb)->type;
230 	const int code = icmp_hdr(skb)->code;
231 	struct tnl_ptk_info tpi;
232 	bool csum_err = false;
233 
234 	if (gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP),
235 			     iph->ihl * 4) < 0) {
236 		if (!csum_err)		/* ignore csum errors. */
237 			return;
238 	}
239 
240 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
241 		ipv4_update_pmtu(skb, dev_net(skb->dev), info,
242 				 skb->dev->ifindex, 0, IPPROTO_GRE, 0);
243 		return;
244 	}
245 	if (type == ICMP_REDIRECT) {
246 		ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex, 0,
247 			      IPPROTO_GRE, 0);
248 		return;
249 	}
250 
251 	ipgre_err(skb, info, &tpi);
252 }
253 
254 static int erspan_rcv(struct sk_buff *skb, struct tnl_ptk_info *tpi,
255 		      int gre_hdr_len)
256 {
257 	struct net *net = dev_net(skb->dev);
258 	struct metadata_dst *tun_dst = NULL;
259 	struct erspan_base_hdr *ershdr;
260 	struct erspan_metadata *pkt_md;
261 	struct ip_tunnel_net *itn;
262 	struct ip_tunnel *tunnel;
263 	const struct iphdr *iph;
264 	struct erspan_md2 *md2;
265 	int ver;
266 	int len;
267 
268 	itn = net_generic(net, erspan_net_id);
269 	len = gre_hdr_len + sizeof(*ershdr);
270 
271 	/* Check based hdr len */
272 	if (unlikely(!pskb_may_pull(skb, len)))
273 		return PACKET_REJECT;
274 
275 	iph = ip_hdr(skb);
276 	ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
277 	ver = ershdr->ver;
278 
279 	/* The original GRE header does not have key field,
280 	 * Use ERSPAN 10-bit session ID as key.
281 	 */
282 	tpi->key = cpu_to_be32(get_session_id(ershdr));
283 	tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex,
284 				  tpi->flags | TUNNEL_KEY,
285 				  iph->saddr, iph->daddr, tpi->key);
286 
287 	if (tunnel) {
288 		len = gre_hdr_len + erspan_hdr_len(ver);
289 		if (unlikely(!pskb_may_pull(skb, len)))
290 			return PACKET_REJECT;
291 
292 		ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
293 		pkt_md = (struct erspan_metadata *)(ershdr + 1);
294 
295 		if (__iptunnel_pull_header(skb,
296 					   len,
297 					   htons(ETH_P_TEB),
298 					   false, false) < 0)
299 			goto drop;
300 
301 		if (tunnel->collect_md) {
302 			struct ip_tunnel_info *info;
303 			struct erspan_metadata *md;
304 			__be64 tun_id;
305 			__be16 flags;
306 
307 			tpi->flags |= TUNNEL_KEY;
308 			flags = tpi->flags;
309 			tun_id = key32_to_tunnel_id(tpi->key);
310 
311 			tun_dst = ip_tun_rx_dst(skb, flags,
312 						tun_id, sizeof(*md));
313 			if (!tun_dst)
314 				return PACKET_REJECT;
315 
316 			md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
317 			md->version = ver;
318 			md2 = &md->u.md2;
319 			memcpy(md2, pkt_md, ver == 1 ? ERSPAN_V1_MDSIZE :
320 						       ERSPAN_V2_MDSIZE);
321 
322 			info = &tun_dst->u.tun_info;
323 			info->key.tun_flags |= TUNNEL_ERSPAN_OPT;
324 			info->options_len = sizeof(*md);
325 		}
326 
327 		skb_reset_mac_header(skb);
328 		ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
329 		return PACKET_RCVD;
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 struct rtable *gre_get_rt(struct sk_buff *skb,
458 				 struct net_device *dev,
459 				 struct flowi4 *fl,
460 				 const struct ip_tunnel_key *key)
461 {
462 	struct net *net = dev_net(dev);
463 
464 	memset(fl, 0, sizeof(*fl));
465 	fl->daddr = key->u.ipv4.dst;
466 	fl->saddr = key->u.ipv4.src;
467 	fl->flowi4_tos = RT_TOS(key->tos);
468 	fl->flowi4_mark = skb->mark;
469 	fl->flowi4_proto = IPPROTO_GRE;
470 
471 	return ip_route_output_key(net, fl);
472 }
473 
474 static struct rtable *prepare_fb_xmit(struct sk_buff *skb,
475 				      struct net_device *dev,
476 				      struct flowi4 *fl,
477 				      int tunnel_hlen)
478 {
479 	struct ip_tunnel_info *tun_info;
480 	const struct ip_tunnel_key *key;
481 	struct rtable *rt = NULL;
482 	int min_headroom;
483 	bool use_cache;
484 	int err;
485 
486 	tun_info = skb_tunnel_info(skb);
487 	key = &tun_info->key;
488 	use_cache = ip_tunnel_dst_cache_usable(skb, tun_info);
489 
490 	if (use_cache)
491 		rt = dst_cache_get_ip4(&tun_info->dst_cache, &fl->saddr);
492 	if (!rt) {
493 		rt = gre_get_rt(skb, dev, fl, key);
494 		if (IS_ERR(rt))
495 			goto err_free_skb;
496 		if (use_cache)
497 			dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst,
498 					  fl->saddr);
499 	}
500 
501 	min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
502 			+ tunnel_hlen + sizeof(struct iphdr);
503 	if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
504 		int head_delta = SKB_DATA_ALIGN(min_headroom -
505 						skb_headroom(skb) +
506 						16);
507 		err = pskb_expand_head(skb, max_t(int, head_delta, 0),
508 				       0, GFP_ATOMIC);
509 		if (unlikely(err))
510 			goto err_free_rt;
511 	}
512 	return rt;
513 
514 err_free_rt:
515 	ip_rt_put(rt);
516 err_free_skb:
517 	kfree_skb(skb);
518 	dev->stats.tx_dropped++;
519 	return NULL;
520 }
521 
522 static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev,
523 			__be16 proto)
524 {
525 	struct ip_tunnel *tunnel = netdev_priv(dev);
526 	struct ip_tunnel_info *tun_info;
527 	const struct ip_tunnel_key *key;
528 	struct rtable *rt = NULL;
529 	struct flowi4 fl;
530 	int tunnel_hlen;
531 	__be16 df, flags;
532 
533 	tun_info = skb_tunnel_info(skb);
534 	if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
535 		     ip_tunnel_info_af(tun_info) != AF_INET))
536 		goto err_free_skb;
537 
538 	key = &tun_info->key;
539 	tunnel_hlen = gre_calc_hlen(key->tun_flags);
540 
541 	rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
542 	if (!rt)
543 		return;
544 
545 	/* Push Tunnel header. */
546 	if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM)))
547 		goto err_free_rt;
548 
549 	flags = tun_info->key.tun_flags &
550 		(TUNNEL_CSUM | TUNNEL_KEY | TUNNEL_SEQ);
551 	gre_build_header(skb, tunnel_hlen, flags, proto,
552 			 tunnel_id_to_key32(tun_info->key.tun_id),
553 			 (flags & TUNNEL_SEQ) ? htonl(tunnel->o_seqno++) : 0);
554 
555 	df = key->tun_flags & TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
556 
557 	iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
558 		      key->tos, key->ttl, df, false);
559 	return;
560 
561 err_free_rt:
562 	ip_rt_put(rt);
563 err_free_skb:
564 	kfree_skb(skb);
565 	dev->stats.tx_dropped++;
566 }
567 
568 static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev,
569 			   __be16 proto)
570 {
571 	struct ip_tunnel *tunnel = netdev_priv(dev);
572 	struct ip_tunnel_info *tun_info;
573 	const struct ip_tunnel_key *key;
574 	struct erspan_metadata *md;
575 	struct rtable *rt = NULL;
576 	bool truncate = false;
577 	struct flowi4 fl;
578 	int tunnel_hlen;
579 	int version;
580 	__be16 df;
581 	int nhoff;
582 	int thoff;
583 
584 	tun_info = skb_tunnel_info(skb);
585 	if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
586 		     ip_tunnel_info_af(tun_info) != AF_INET))
587 		goto err_free_skb;
588 
589 	key = &tun_info->key;
590 	md = ip_tunnel_info_opts(tun_info);
591 	if (!md)
592 		goto err_free_rt;
593 
594 	/* ERSPAN has fixed 8 byte GRE header */
595 	version = md->version;
596 	tunnel_hlen = 8 + erspan_hdr_len(version);
597 
598 	rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
599 	if (!rt)
600 		return;
601 
602 	if (gre_handle_offloads(skb, false))
603 		goto err_free_rt;
604 
605 	if (skb->len > dev->mtu + dev->hard_header_len) {
606 		pskb_trim(skb, dev->mtu + dev->hard_header_len);
607 		truncate = true;
608 	}
609 
610 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
611 	if (skb->protocol == htons(ETH_P_IP) &&
612 	    (ntohs(ip_hdr(skb)->tot_len) > skb->len - nhoff))
613 		truncate = true;
614 
615 	thoff = skb_transport_header(skb) - skb_mac_header(skb);
616 	if (skb->protocol == htons(ETH_P_IPV6) &&
617 	    (ntohs(ipv6_hdr(skb)->payload_len) > skb->len - thoff))
618 		truncate = true;
619 
620 	if (version == 1) {
621 		erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)),
622 				    ntohl(md->u.index), truncate, true);
623 	} else if (version == 2) {
624 		erspan_build_header_v2(skb,
625 				       ntohl(tunnel_id_to_key32(key->tun_id)),
626 				       md->u.md2.dir,
627 				       get_hwid(&md->u.md2),
628 				       truncate, true);
629 	} else {
630 		goto err_free_rt;
631 	}
632 
633 	gre_build_header(skb, 8, TUNNEL_SEQ,
634 			 htons(ETH_P_ERSPAN), 0, htonl(tunnel->o_seqno++));
635 
636 	df = key->tun_flags & TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
637 
638 	iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
639 		      key->tos, key->ttl, df, false);
640 	return;
641 
642 err_free_rt:
643 	ip_rt_put(rt);
644 err_free_skb:
645 	kfree_skb(skb);
646 	dev->stats.tx_dropped++;
647 }
648 
649 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
650 {
651 	struct ip_tunnel_info *info = skb_tunnel_info(skb);
652 	struct rtable *rt;
653 	struct flowi4 fl4;
654 
655 	if (ip_tunnel_info_af(info) != AF_INET)
656 		return -EINVAL;
657 
658 	rt = gre_get_rt(skb, dev, &fl4, &info->key);
659 	if (IS_ERR(rt))
660 		return PTR_ERR(rt);
661 
662 	ip_rt_put(rt);
663 	info->key.u.ipv4.src = fl4.saddr;
664 	return 0;
665 }
666 
667 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
668 			      struct net_device *dev)
669 {
670 	struct ip_tunnel *tunnel = netdev_priv(dev);
671 	const struct iphdr *tnl_params;
672 
673 	if (tunnel->collect_md) {
674 		gre_fb_xmit(skb, dev, skb->protocol);
675 		return NETDEV_TX_OK;
676 	}
677 
678 	if (dev->header_ops) {
679 		/* Need space for new headers */
680 		if (skb_cow_head(skb, dev->needed_headroom -
681 				      (tunnel->hlen + sizeof(struct iphdr))))
682 			goto free_skb;
683 
684 		tnl_params = (const struct iphdr *)skb->data;
685 
686 		/* Pull skb since ip_tunnel_xmit() needs skb->data pointing
687 		 * to gre header.
688 		 */
689 		skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
690 		skb_reset_mac_header(skb);
691 	} else {
692 		if (skb_cow_head(skb, dev->needed_headroom))
693 			goto free_skb;
694 
695 		tnl_params = &tunnel->parms.iph;
696 	}
697 
698 	if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
699 		goto free_skb;
700 
701 	__gre_xmit(skb, dev, tnl_params, skb->protocol);
702 	return NETDEV_TX_OK;
703 
704 free_skb:
705 	kfree_skb(skb);
706 	dev->stats.tx_dropped++;
707 	return NETDEV_TX_OK;
708 }
709 
710 static netdev_tx_t erspan_xmit(struct sk_buff *skb,
711 			       struct net_device *dev)
712 {
713 	struct ip_tunnel *tunnel = netdev_priv(dev);
714 	bool truncate = false;
715 
716 	if (tunnel->collect_md) {
717 		erspan_fb_xmit(skb, dev, skb->protocol);
718 		return NETDEV_TX_OK;
719 	}
720 
721 	if (gre_handle_offloads(skb, false))
722 		goto free_skb;
723 
724 	if (skb_cow_head(skb, dev->needed_headroom))
725 		goto free_skb;
726 
727 	if (skb->len > dev->mtu + dev->hard_header_len) {
728 		pskb_trim(skb, dev->mtu + dev->hard_header_len);
729 		truncate = true;
730 	}
731 
732 	/* Push ERSPAN header */
733 	if (tunnel->erspan_ver == 1)
734 		erspan_build_header(skb, ntohl(tunnel->parms.o_key),
735 				    tunnel->index,
736 				    truncate, true);
737 	else
738 		erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key),
739 				       tunnel->dir, tunnel->hwid,
740 				       truncate, true);
741 
742 	tunnel->parms.o_flags &= ~TUNNEL_KEY;
743 	__gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_ERSPAN));
744 	return NETDEV_TX_OK;
745 
746 free_skb:
747 	kfree_skb(skb);
748 	dev->stats.tx_dropped++;
749 	return NETDEV_TX_OK;
750 }
751 
752 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
753 				struct net_device *dev)
754 {
755 	struct ip_tunnel *tunnel = netdev_priv(dev);
756 
757 	if (tunnel->collect_md) {
758 		gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
759 		return NETDEV_TX_OK;
760 	}
761 
762 	if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
763 		goto free_skb;
764 
765 	if (skb_cow_head(skb, dev->needed_headroom))
766 		goto free_skb;
767 
768 	__gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
769 	return NETDEV_TX_OK;
770 
771 free_skb:
772 	kfree_skb(skb);
773 	dev->stats.tx_dropped++;
774 	return NETDEV_TX_OK;
775 }
776 
777 static void ipgre_link_update(struct net_device *dev, bool set_mtu)
778 {
779 	struct ip_tunnel *tunnel = netdev_priv(dev);
780 	int len;
781 
782 	len = tunnel->tun_hlen;
783 	tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
784 	len = tunnel->tun_hlen - len;
785 	tunnel->hlen = tunnel->hlen + len;
786 
787 	dev->needed_headroom = dev->needed_headroom + len;
788 	if (set_mtu)
789 		dev->mtu = max_t(int, dev->mtu - len, 68);
790 
791 	if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
792 		if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
793 		    tunnel->encap.type == TUNNEL_ENCAP_NONE) {
794 			dev->features |= NETIF_F_GSO_SOFTWARE;
795 			dev->hw_features |= NETIF_F_GSO_SOFTWARE;
796 		} else {
797 			dev->features &= ~NETIF_F_GSO_SOFTWARE;
798 			dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
799 		}
800 		dev->features |= NETIF_F_LLTX;
801 	} else {
802 		dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
803 		dev->features &= ~(NETIF_F_LLTX | NETIF_F_GSO_SOFTWARE);
804 	}
805 }
806 
807 static int ipgre_tunnel_ioctl(struct net_device *dev,
808 			      struct ifreq *ifr, int cmd)
809 {
810 	struct ip_tunnel_parm p;
811 	int err;
812 
813 	if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
814 		return -EFAULT;
815 
816 	if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
817 		if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
818 		    p.iph.ihl != 5 || (p.iph.frag_off & htons(~IP_DF)) ||
819 		    ((p.i_flags | p.o_flags) & (GRE_VERSION | GRE_ROUTING)))
820 			return -EINVAL;
821 	}
822 
823 	p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
824 	p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
825 
826 	err = ip_tunnel_ioctl(dev, &p, cmd);
827 	if (err)
828 		return err;
829 
830 	if (cmd == SIOCCHGTUNNEL) {
831 		struct ip_tunnel *t = netdev_priv(dev);
832 
833 		t->parms.i_flags = p.i_flags;
834 		t->parms.o_flags = p.o_flags;
835 
836 		if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
837 			ipgre_link_update(dev, true);
838 	}
839 
840 	p.i_flags = gre_tnl_flags_to_gre_flags(p.i_flags);
841 	p.o_flags = gre_tnl_flags_to_gre_flags(p.o_flags);
842 
843 	if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
844 		return -EFAULT;
845 
846 	return 0;
847 }
848 
849 /* Nice toy. Unfortunately, useless in real life :-)
850    It allows to construct virtual multiprotocol broadcast "LAN"
851    over the Internet, provided multicast routing is tuned.
852 
853 
854    I have no idea was this bicycle invented before me,
855    so that I had to set ARPHRD_IPGRE to a random value.
856    I have an impression, that Cisco could make something similar,
857    but this feature is apparently missing in IOS<=11.2(8).
858 
859    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
860    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
861 
862    ping -t 255 224.66.66.66
863 
864    If nobody answers, mbone does not work.
865 
866    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
867    ip addr add 10.66.66.<somewhat>/24 dev Universe
868    ifconfig Universe up
869    ifconfig Universe add fe80::<Your_real_addr>/10
870    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
871    ftp 10.66.66.66
872    ...
873    ftp fec0:6666:6666::193.233.7.65
874    ...
875  */
876 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
877 			unsigned short type,
878 			const void *daddr, const void *saddr, unsigned int len)
879 {
880 	struct ip_tunnel *t = netdev_priv(dev);
881 	struct iphdr *iph;
882 	struct gre_base_hdr *greh;
883 
884 	iph = skb_push(skb, t->hlen + sizeof(*iph));
885 	greh = (struct gre_base_hdr *)(iph+1);
886 	greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
887 	greh->protocol = htons(type);
888 
889 	memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
890 
891 	/* Set the source hardware address. */
892 	if (saddr)
893 		memcpy(&iph->saddr, saddr, 4);
894 	if (daddr)
895 		memcpy(&iph->daddr, daddr, 4);
896 	if (iph->daddr)
897 		return t->hlen + sizeof(*iph);
898 
899 	return -(t->hlen + sizeof(*iph));
900 }
901 
902 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
903 {
904 	const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
905 	memcpy(haddr, &iph->saddr, 4);
906 	return 4;
907 }
908 
909 static const struct header_ops ipgre_header_ops = {
910 	.create	= ipgre_header,
911 	.parse	= ipgre_header_parse,
912 };
913 
914 #ifdef CONFIG_NET_IPGRE_BROADCAST
915 static int ipgre_open(struct net_device *dev)
916 {
917 	struct ip_tunnel *t = netdev_priv(dev);
918 
919 	if (ipv4_is_multicast(t->parms.iph.daddr)) {
920 		struct flowi4 fl4;
921 		struct rtable *rt;
922 
923 		rt = ip_route_output_gre(t->net, &fl4,
924 					 t->parms.iph.daddr,
925 					 t->parms.iph.saddr,
926 					 t->parms.o_key,
927 					 RT_TOS(t->parms.iph.tos),
928 					 t->parms.link);
929 		if (IS_ERR(rt))
930 			return -EADDRNOTAVAIL;
931 		dev = rt->dst.dev;
932 		ip_rt_put(rt);
933 		if (!__in_dev_get_rtnl(dev))
934 			return -EADDRNOTAVAIL;
935 		t->mlink = dev->ifindex;
936 		ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
937 	}
938 	return 0;
939 }
940 
941 static int ipgre_close(struct net_device *dev)
942 {
943 	struct ip_tunnel *t = netdev_priv(dev);
944 
945 	if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
946 		struct in_device *in_dev;
947 		in_dev = inetdev_by_index(t->net, t->mlink);
948 		if (in_dev)
949 			ip_mc_dec_group(in_dev, t->parms.iph.daddr);
950 	}
951 	return 0;
952 }
953 #endif
954 
955 static const struct net_device_ops ipgre_netdev_ops = {
956 	.ndo_init		= ipgre_tunnel_init,
957 	.ndo_uninit		= ip_tunnel_uninit,
958 #ifdef CONFIG_NET_IPGRE_BROADCAST
959 	.ndo_open		= ipgre_open,
960 	.ndo_stop		= ipgre_close,
961 #endif
962 	.ndo_start_xmit		= ipgre_xmit,
963 	.ndo_do_ioctl		= ipgre_tunnel_ioctl,
964 	.ndo_change_mtu		= ip_tunnel_change_mtu,
965 	.ndo_get_stats64	= ip_tunnel_get_stats64,
966 	.ndo_get_iflink		= ip_tunnel_get_iflink,
967 };
968 
969 #define GRE_FEATURES (NETIF_F_SG |		\
970 		      NETIF_F_FRAGLIST |	\
971 		      NETIF_F_HIGHDMA |		\
972 		      NETIF_F_HW_CSUM)
973 
974 static void ipgre_tunnel_setup(struct net_device *dev)
975 {
976 	dev->netdev_ops		= &ipgre_netdev_ops;
977 	dev->type		= ARPHRD_IPGRE;
978 	ip_tunnel_setup(dev, ipgre_net_id);
979 }
980 
981 static void __gre_tunnel_init(struct net_device *dev)
982 {
983 	struct ip_tunnel *tunnel;
984 	int t_hlen;
985 
986 	tunnel = netdev_priv(dev);
987 	tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
988 	tunnel->parms.iph.protocol = IPPROTO_GRE;
989 
990 	tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
991 
992 	t_hlen = tunnel->hlen + sizeof(struct iphdr);
993 
994 	dev->features		|= GRE_FEATURES;
995 	dev->hw_features	|= GRE_FEATURES;
996 
997 	if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
998 		/* TCP offload with GRE SEQ is not supported, nor
999 		 * can we support 2 levels of outer headers requiring
1000 		 * an update.
1001 		 */
1002 		if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
1003 		    (tunnel->encap.type == TUNNEL_ENCAP_NONE)) {
1004 			dev->features    |= NETIF_F_GSO_SOFTWARE;
1005 			dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1006 		}
1007 
1008 		/* Can use a lockless transmit, unless we generate
1009 		 * output sequences
1010 		 */
1011 		dev->features |= NETIF_F_LLTX;
1012 	}
1013 }
1014 
1015 static int ipgre_tunnel_init(struct net_device *dev)
1016 {
1017 	struct ip_tunnel *tunnel = netdev_priv(dev);
1018 	struct iphdr *iph = &tunnel->parms.iph;
1019 
1020 	__gre_tunnel_init(dev);
1021 
1022 	memcpy(dev->dev_addr, &iph->saddr, 4);
1023 	memcpy(dev->broadcast, &iph->daddr, 4);
1024 
1025 	dev->flags		= IFF_NOARP;
1026 	netif_keep_dst(dev);
1027 	dev->addr_len		= 4;
1028 
1029 	if (iph->daddr && !tunnel->collect_md) {
1030 #ifdef CONFIG_NET_IPGRE_BROADCAST
1031 		if (ipv4_is_multicast(iph->daddr)) {
1032 			if (!iph->saddr)
1033 				return -EINVAL;
1034 			dev->flags = IFF_BROADCAST;
1035 			dev->header_ops = &ipgre_header_ops;
1036 		}
1037 #endif
1038 	} else if (!tunnel->collect_md) {
1039 		dev->header_ops = &ipgre_header_ops;
1040 	}
1041 
1042 	return ip_tunnel_init(dev);
1043 }
1044 
1045 static const struct gre_protocol ipgre_protocol = {
1046 	.handler     = gre_rcv,
1047 	.err_handler = gre_err,
1048 };
1049 
1050 static int __net_init ipgre_init_net(struct net *net)
1051 {
1052 	return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
1053 }
1054 
1055 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net)
1056 {
1057 	ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops);
1058 }
1059 
1060 static struct pernet_operations ipgre_net_ops = {
1061 	.init = ipgre_init_net,
1062 	.exit_batch = ipgre_exit_batch_net,
1063 	.id   = &ipgre_net_id,
1064 	.size = sizeof(struct ip_tunnel_net),
1065 };
1066 
1067 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1068 				 struct netlink_ext_ack *extack)
1069 {
1070 	__be16 flags;
1071 
1072 	if (!data)
1073 		return 0;
1074 
1075 	flags = 0;
1076 	if (data[IFLA_GRE_IFLAGS])
1077 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1078 	if (data[IFLA_GRE_OFLAGS])
1079 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1080 	if (flags & (GRE_VERSION|GRE_ROUTING))
1081 		return -EINVAL;
1082 
1083 	if (data[IFLA_GRE_COLLECT_METADATA] &&
1084 	    data[IFLA_GRE_ENCAP_TYPE] &&
1085 	    nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1086 		return -EINVAL;
1087 
1088 	return 0;
1089 }
1090 
1091 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1092 			      struct netlink_ext_ack *extack)
1093 {
1094 	__be32 daddr;
1095 
1096 	if (tb[IFLA_ADDRESS]) {
1097 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1098 			return -EINVAL;
1099 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1100 			return -EADDRNOTAVAIL;
1101 	}
1102 
1103 	if (!data)
1104 		goto out;
1105 
1106 	if (data[IFLA_GRE_REMOTE]) {
1107 		memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1108 		if (!daddr)
1109 			return -EINVAL;
1110 	}
1111 
1112 out:
1113 	return ipgre_tunnel_validate(tb, data, extack);
1114 }
1115 
1116 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1117 			   struct netlink_ext_ack *extack)
1118 {
1119 	__be16 flags = 0;
1120 	int ret;
1121 
1122 	if (!data)
1123 		return 0;
1124 
1125 	ret = ipgre_tap_validate(tb, data, extack);
1126 	if (ret)
1127 		return ret;
1128 
1129 	/* ERSPAN should only have GRE sequence and key flag */
1130 	if (data[IFLA_GRE_OFLAGS])
1131 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1132 	if (data[IFLA_GRE_IFLAGS])
1133 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1134 	if (!data[IFLA_GRE_COLLECT_METADATA] &&
1135 	    flags != (GRE_SEQ | GRE_KEY))
1136 		return -EINVAL;
1137 
1138 	/* ERSPAN Session ID only has 10-bit. Since we reuse
1139 	 * 32-bit key field as ID, check it's range.
1140 	 */
1141 	if (data[IFLA_GRE_IKEY] &&
1142 	    (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1143 		return -EINVAL;
1144 
1145 	if (data[IFLA_GRE_OKEY] &&
1146 	    (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1147 		return -EINVAL;
1148 
1149 	return 0;
1150 }
1151 
1152 static int ipgre_netlink_parms(struct net_device *dev,
1153 				struct nlattr *data[],
1154 				struct nlattr *tb[],
1155 				struct ip_tunnel_parm *parms,
1156 				__u32 *fwmark)
1157 {
1158 	struct ip_tunnel *t = netdev_priv(dev);
1159 
1160 	memset(parms, 0, sizeof(*parms));
1161 
1162 	parms->iph.protocol = IPPROTO_GRE;
1163 
1164 	if (!data)
1165 		return 0;
1166 
1167 	if (data[IFLA_GRE_LINK])
1168 		parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1169 
1170 	if (data[IFLA_GRE_IFLAGS])
1171 		parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
1172 
1173 	if (data[IFLA_GRE_OFLAGS])
1174 		parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
1175 
1176 	if (data[IFLA_GRE_IKEY])
1177 		parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1178 
1179 	if (data[IFLA_GRE_OKEY])
1180 		parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1181 
1182 	if (data[IFLA_GRE_LOCAL])
1183 		parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1184 
1185 	if (data[IFLA_GRE_REMOTE])
1186 		parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1187 
1188 	if (data[IFLA_GRE_TTL])
1189 		parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1190 
1191 	if (data[IFLA_GRE_TOS])
1192 		parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1193 
1194 	if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1195 		if (t->ignore_df)
1196 			return -EINVAL;
1197 		parms->iph.frag_off = htons(IP_DF);
1198 	}
1199 
1200 	if (data[IFLA_GRE_COLLECT_METADATA]) {
1201 		t->collect_md = true;
1202 		if (dev->type == ARPHRD_IPGRE)
1203 			dev->type = ARPHRD_NONE;
1204 	}
1205 
1206 	if (data[IFLA_GRE_IGNORE_DF]) {
1207 		if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1208 		  && (parms->iph.frag_off & htons(IP_DF)))
1209 			return -EINVAL;
1210 		t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1211 	}
1212 
1213 	if (data[IFLA_GRE_FWMARK])
1214 		*fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1215 
1216 	if (data[IFLA_GRE_ERSPAN_VER]) {
1217 		t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]);
1218 
1219 		if (t->erspan_ver != 1 && t->erspan_ver != 2)
1220 			return -EINVAL;
1221 	}
1222 
1223 	if (t->erspan_ver == 1) {
1224 		if (data[IFLA_GRE_ERSPAN_INDEX]) {
1225 			t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1226 			if (t->index & ~INDEX_MASK)
1227 				return -EINVAL;
1228 		}
1229 	} else if (t->erspan_ver == 2) {
1230 		if (data[IFLA_GRE_ERSPAN_DIR]) {
1231 			t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]);
1232 			if (t->dir & ~(DIR_MASK >> DIR_OFFSET))
1233 				return -EINVAL;
1234 		}
1235 		if (data[IFLA_GRE_ERSPAN_HWID]) {
1236 			t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]);
1237 			if (t->hwid & ~(HWID_MASK >> HWID_OFFSET))
1238 				return -EINVAL;
1239 		}
1240 	}
1241 
1242 	return 0;
1243 }
1244 
1245 /* This function returns true when ENCAP attributes are present in the nl msg */
1246 static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1247 				      struct ip_tunnel_encap *ipencap)
1248 {
1249 	bool ret = false;
1250 
1251 	memset(ipencap, 0, sizeof(*ipencap));
1252 
1253 	if (!data)
1254 		return ret;
1255 
1256 	if (data[IFLA_GRE_ENCAP_TYPE]) {
1257 		ret = true;
1258 		ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1259 	}
1260 
1261 	if (data[IFLA_GRE_ENCAP_FLAGS]) {
1262 		ret = true;
1263 		ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1264 	}
1265 
1266 	if (data[IFLA_GRE_ENCAP_SPORT]) {
1267 		ret = true;
1268 		ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1269 	}
1270 
1271 	if (data[IFLA_GRE_ENCAP_DPORT]) {
1272 		ret = true;
1273 		ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1274 	}
1275 
1276 	return ret;
1277 }
1278 
1279 static int gre_tap_init(struct net_device *dev)
1280 {
1281 	__gre_tunnel_init(dev);
1282 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1283 	netif_keep_dst(dev);
1284 
1285 	return ip_tunnel_init(dev);
1286 }
1287 
1288 static const struct net_device_ops gre_tap_netdev_ops = {
1289 	.ndo_init		= gre_tap_init,
1290 	.ndo_uninit		= ip_tunnel_uninit,
1291 	.ndo_start_xmit		= gre_tap_xmit,
1292 	.ndo_set_mac_address 	= eth_mac_addr,
1293 	.ndo_validate_addr	= eth_validate_addr,
1294 	.ndo_change_mtu		= ip_tunnel_change_mtu,
1295 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1296 	.ndo_get_iflink		= ip_tunnel_get_iflink,
1297 	.ndo_fill_metadata_dst	= gre_fill_metadata_dst,
1298 };
1299 
1300 static int erspan_tunnel_init(struct net_device *dev)
1301 {
1302 	struct ip_tunnel *tunnel = netdev_priv(dev);
1303 	int t_hlen;
1304 
1305 	tunnel->tun_hlen = 8;
1306 	tunnel->parms.iph.protocol = IPPROTO_GRE;
1307 	tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1308 		       erspan_hdr_len(tunnel->erspan_ver);
1309 	t_hlen = tunnel->hlen + sizeof(struct iphdr);
1310 
1311 	dev->features		|= GRE_FEATURES;
1312 	dev->hw_features	|= GRE_FEATURES;
1313 	dev->priv_flags		|= IFF_LIVE_ADDR_CHANGE;
1314 	netif_keep_dst(dev);
1315 
1316 	return ip_tunnel_init(dev);
1317 }
1318 
1319 static const struct net_device_ops erspan_netdev_ops = {
1320 	.ndo_init		= erspan_tunnel_init,
1321 	.ndo_uninit		= ip_tunnel_uninit,
1322 	.ndo_start_xmit		= erspan_xmit,
1323 	.ndo_set_mac_address	= eth_mac_addr,
1324 	.ndo_validate_addr	= eth_validate_addr,
1325 	.ndo_change_mtu		= ip_tunnel_change_mtu,
1326 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1327 	.ndo_get_iflink		= ip_tunnel_get_iflink,
1328 	.ndo_fill_metadata_dst	= gre_fill_metadata_dst,
1329 };
1330 
1331 static void ipgre_tap_setup(struct net_device *dev)
1332 {
1333 	ether_setup(dev);
1334 	dev->max_mtu = 0;
1335 	dev->netdev_ops	= &gre_tap_netdev_ops;
1336 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1337 	dev->priv_flags	|= IFF_LIVE_ADDR_CHANGE;
1338 	ip_tunnel_setup(dev, gre_tap_net_id);
1339 }
1340 
1341 bool is_gretap_dev(const struct net_device *dev)
1342 {
1343 	return dev->netdev_ops == &gre_tap_netdev_ops;
1344 }
1345 EXPORT_SYMBOL_GPL(is_gretap_dev);
1346 
1347 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1348 			 struct nlattr *tb[], struct nlattr *data[],
1349 			 struct netlink_ext_ack *extack)
1350 {
1351 	struct ip_tunnel_parm p;
1352 	struct ip_tunnel_encap ipencap;
1353 	__u32 fwmark = 0;
1354 	int err;
1355 
1356 	if (ipgre_netlink_encap_parms(data, &ipencap)) {
1357 		struct ip_tunnel *t = netdev_priv(dev);
1358 		err = ip_tunnel_encap_setup(t, &ipencap);
1359 
1360 		if (err < 0)
1361 			return err;
1362 	}
1363 
1364 	err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1365 	if (err < 0)
1366 		return err;
1367 	return ip_tunnel_newlink(dev, tb, &p, fwmark);
1368 }
1369 
1370 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1371 			    struct nlattr *data[],
1372 			    struct netlink_ext_ack *extack)
1373 {
1374 	struct ip_tunnel *t = netdev_priv(dev);
1375 	struct ip_tunnel_encap ipencap;
1376 	__u32 fwmark = t->fwmark;
1377 	struct ip_tunnel_parm p;
1378 	int err;
1379 
1380 	if (ipgre_netlink_encap_parms(data, &ipencap)) {
1381 		err = ip_tunnel_encap_setup(t, &ipencap);
1382 
1383 		if (err < 0)
1384 			return err;
1385 	}
1386 
1387 	err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1388 	if (err < 0)
1389 		return err;
1390 
1391 	err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1392 	if (err < 0)
1393 		return err;
1394 
1395 	t->parms.i_flags = p.i_flags;
1396 	t->parms.o_flags = p.o_flags;
1397 
1398 	if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
1399 		ipgre_link_update(dev, !tb[IFLA_MTU]);
1400 
1401 	return 0;
1402 }
1403 
1404 static size_t ipgre_get_size(const struct net_device *dev)
1405 {
1406 	return
1407 		/* IFLA_GRE_LINK */
1408 		nla_total_size(4) +
1409 		/* IFLA_GRE_IFLAGS */
1410 		nla_total_size(2) +
1411 		/* IFLA_GRE_OFLAGS */
1412 		nla_total_size(2) +
1413 		/* IFLA_GRE_IKEY */
1414 		nla_total_size(4) +
1415 		/* IFLA_GRE_OKEY */
1416 		nla_total_size(4) +
1417 		/* IFLA_GRE_LOCAL */
1418 		nla_total_size(4) +
1419 		/* IFLA_GRE_REMOTE */
1420 		nla_total_size(4) +
1421 		/* IFLA_GRE_TTL */
1422 		nla_total_size(1) +
1423 		/* IFLA_GRE_TOS */
1424 		nla_total_size(1) +
1425 		/* IFLA_GRE_PMTUDISC */
1426 		nla_total_size(1) +
1427 		/* IFLA_GRE_ENCAP_TYPE */
1428 		nla_total_size(2) +
1429 		/* IFLA_GRE_ENCAP_FLAGS */
1430 		nla_total_size(2) +
1431 		/* IFLA_GRE_ENCAP_SPORT */
1432 		nla_total_size(2) +
1433 		/* IFLA_GRE_ENCAP_DPORT */
1434 		nla_total_size(2) +
1435 		/* IFLA_GRE_COLLECT_METADATA */
1436 		nla_total_size(0) +
1437 		/* IFLA_GRE_IGNORE_DF */
1438 		nla_total_size(1) +
1439 		/* IFLA_GRE_FWMARK */
1440 		nla_total_size(4) +
1441 		/* IFLA_GRE_ERSPAN_INDEX */
1442 		nla_total_size(4) +
1443 		/* IFLA_GRE_ERSPAN_VER */
1444 		nla_total_size(1) +
1445 		/* IFLA_GRE_ERSPAN_DIR */
1446 		nla_total_size(1) +
1447 		/* IFLA_GRE_ERSPAN_HWID */
1448 		nla_total_size(2) +
1449 		0;
1450 }
1451 
1452 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1453 {
1454 	struct ip_tunnel *t = netdev_priv(dev);
1455 	struct ip_tunnel_parm *p = &t->parms;
1456 
1457 	if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1458 	    nla_put_be16(skb, IFLA_GRE_IFLAGS,
1459 			 gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1460 	    nla_put_be16(skb, IFLA_GRE_OFLAGS,
1461 			 gre_tnl_flags_to_gre_flags(p->o_flags)) ||
1462 	    nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1463 	    nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1464 	    nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1465 	    nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1466 	    nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1467 	    nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1468 	    nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1469 		       !!(p->iph.frag_off & htons(IP_DF))) ||
1470 	    nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1471 		goto nla_put_failure;
1472 
1473 	if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1474 			t->encap.type) ||
1475 	    nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1476 			 t->encap.sport) ||
1477 	    nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1478 			 t->encap.dport) ||
1479 	    nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1480 			t->encap.flags))
1481 		goto nla_put_failure;
1482 
1483 	if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1484 		goto nla_put_failure;
1485 
1486 	if (t->collect_md) {
1487 		if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1488 			goto nla_put_failure;
1489 	}
1490 
1491 	if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1492 		goto nla_put_failure;
1493 
1494 	if (t->erspan_ver == 1) {
1495 		if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1496 			goto nla_put_failure;
1497 	} else if (t->erspan_ver == 2) {
1498 		if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1499 			goto nla_put_failure;
1500 		if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1501 			goto nla_put_failure;
1502 	}
1503 
1504 	return 0;
1505 
1506 nla_put_failure:
1507 	return -EMSGSIZE;
1508 }
1509 
1510 static void erspan_setup(struct net_device *dev)
1511 {
1512 	ether_setup(dev);
1513 	dev->netdev_ops = &erspan_netdev_ops;
1514 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1515 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1516 	ip_tunnel_setup(dev, erspan_net_id);
1517 }
1518 
1519 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1520 	[IFLA_GRE_LINK]		= { .type = NLA_U32 },
1521 	[IFLA_GRE_IFLAGS]	= { .type = NLA_U16 },
1522 	[IFLA_GRE_OFLAGS]	= { .type = NLA_U16 },
1523 	[IFLA_GRE_IKEY]		= { .type = NLA_U32 },
1524 	[IFLA_GRE_OKEY]		= { .type = NLA_U32 },
1525 	[IFLA_GRE_LOCAL]	= { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1526 	[IFLA_GRE_REMOTE]	= { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1527 	[IFLA_GRE_TTL]		= { .type = NLA_U8 },
1528 	[IFLA_GRE_TOS]		= { .type = NLA_U8 },
1529 	[IFLA_GRE_PMTUDISC]	= { .type = NLA_U8 },
1530 	[IFLA_GRE_ENCAP_TYPE]	= { .type = NLA_U16 },
1531 	[IFLA_GRE_ENCAP_FLAGS]	= { .type = NLA_U16 },
1532 	[IFLA_GRE_ENCAP_SPORT]	= { .type = NLA_U16 },
1533 	[IFLA_GRE_ENCAP_DPORT]	= { .type = NLA_U16 },
1534 	[IFLA_GRE_COLLECT_METADATA]	= { .type = NLA_FLAG },
1535 	[IFLA_GRE_IGNORE_DF]	= { .type = NLA_U8 },
1536 	[IFLA_GRE_FWMARK]	= { .type = NLA_U32 },
1537 	[IFLA_GRE_ERSPAN_INDEX]	= { .type = NLA_U32 },
1538 	[IFLA_GRE_ERSPAN_VER]	= { .type = NLA_U8 },
1539 	[IFLA_GRE_ERSPAN_DIR]	= { .type = NLA_U8 },
1540 	[IFLA_GRE_ERSPAN_HWID]	= { .type = NLA_U16 },
1541 };
1542 
1543 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1544 	.kind		= "gre",
1545 	.maxtype	= IFLA_GRE_MAX,
1546 	.policy		= ipgre_policy,
1547 	.priv_size	= sizeof(struct ip_tunnel),
1548 	.setup		= ipgre_tunnel_setup,
1549 	.validate	= ipgre_tunnel_validate,
1550 	.newlink	= ipgre_newlink,
1551 	.changelink	= ipgre_changelink,
1552 	.dellink	= ip_tunnel_dellink,
1553 	.get_size	= ipgre_get_size,
1554 	.fill_info	= ipgre_fill_info,
1555 	.get_link_net	= ip_tunnel_get_link_net,
1556 };
1557 
1558 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1559 	.kind		= "gretap",
1560 	.maxtype	= IFLA_GRE_MAX,
1561 	.policy		= ipgre_policy,
1562 	.priv_size	= sizeof(struct ip_tunnel),
1563 	.setup		= ipgre_tap_setup,
1564 	.validate	= ipgre_tap_validate,
1565 	.newlink	= ipgre_newlink,
1566 	.changelink	= ipgre_changelink,
1567 	.dellink	= ip_tunnel_dellink,
1568 	.get_size	= ipgre_get_size,
1569 	.fill_info	= ipgre_fill_info,
1570 	.get_link_net	= ip_tunnel_get_link_net,
1571 };
1572 
1573 static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1574 	.kind		= "erspan",
1575 	.maxtype	= IFLA_GRE_MAX,
1576 	.policy		= ipgre_policy,
1577 	.priv_size	= sizeof(struct ip_tunnel),
1578 	.setup		= erspan_setup,
1579 	.validate	= erspan_validate,
1580 	.newlink	= ipgre_newlink,
1581 	.changelink	= ipgre_changelink,
1582 	.dellink	= ip_tunnel_dellink,
1583 	.get_size	= ipgre_get_size,
1584 	.fill_info	= ipgre_fill_info,
1585 	.get_link_net	= ip_tunnel_get_link_net,
1586 };
1587 
1588 struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1589 					u8 name_assign_type)
1590 {
1591 	struct nlattr *tb[IFLA_MAX + 1];
1592 	struct net_device *dev;
1593 	LIST_HEAD(list_kill);
1594 	struct ip_tunnel *t;
1595 	int err;
1596 
1597 	memset(&tb, 0, sizeof(tb));
1598 
1599 	dev = rtnl_create_link(net, name, name_assign_type,
1600 			       &ipgre_tap_ops, tb);
1601 	if (IS_ERR(dev))
1602 		return dev;
1603 
1604 	/* Configure flow based GRE device. */
1605 	t = netdev_priv(dev);
1606 	t->collect_md = true;
1607 
1608 	err = ipgre_newlink(net, dev, tb, NULL, NULL);
1609 	if (err < 0) {
1610 		free_netdev(dev);
1611 		return ERR_PTR(err);
1612 	}
1613 
1614 	/* openvswitch users expect packet sizes to be unrestricted,
1615 	 * so set the largest MTU we can.
1616 	 */
1617 	err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1618 	if (err)
1619 		goto out;
1620 
1621 	err = rtnl_configure_link(dev, NULL);
1622 	if (err < 0)
1623 		goto out;
1624 
1625 	return dev;
1626 out:
1627 	ip_tunnel_dellink(dev, &list_kill);
1628 	unregister_netdevice_many(&list_kill);
1629 	return ERR_PTR(err);
1630 }
1631 EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1632 
1633 static int __net_init ipgre_tap_init_net(struct net *net)
1634 {
1635 	return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1636 }
1637 
1638 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net)
1639 {
1640 	ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops);
1641 }
1642 
1643 static struct pernet_operations ipgre_tap_net_ops = {
1644 	.init = ipgre_tap_init_net,
1645 	.exit_batch = ipgre_tap_exit_batch_net,
1646 	.id   = &gre_tap_net_id,
1647 	.size = sizeof(struct ip_tunnel_net),
1648 };
1649 
1650 static int __net_init erspan_init_net(struct net *net)
1651 {
1652 	return ip_tunnel_init_net(net, erspan_net_id,
1653 				  &erspan_link_ops, "erspan0");
1654 }
1655 
1656 static void __net_exit erspan_exit_batch_net(struct list_head *net_list)
1657 {
1658 	ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops);
1659 }
1660 
1661 static struct pernet_operations erspan_net_ops = {
1662 	.init = erspan_init_net,
1663 	.exit_batch = erspan_exit_batch_net,
1664 	.id   = &erspan_net_id,
1665 	.size = sizeof(struct ip_tunnel_net),
1666 };
1667 
1668 static int __init ipgre_init(void)
1669 {
1670 	int err;
1671 
1672 	pr_info("GRE over IPv4 tunneling driver\n");
1673 
1674 	err = register_pernet_device(&ipgre_net_ops);
1675 	if (err < 0)
1676 		return err;
1677 
1678 	err = register_pernet_device(&ipgre_tap_net_ops);
1679 	if (err < 0)
1680 		goto pnet_tap_failed;
1681 
1682 	err = register_pernet_device(&erspan_net_ops);
1683 	if (err < 0)
1684 		goto pnet_erspan_failed;
1685 
1686 	err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1687 	if (err < 0) {
1688 		pr_info("%s: can't add protocol\n", __func__);
1689 		goto add_proto_failed;
1690 	}
1691 
1692 	err = rtnl_link_register(&ipgre_link_ops);
1693 	if (err < 0)
1694 		goto rtnl_link_failed;
1695 
1696 	err = rtnl_link_register(&ipgre_tap_ops);
1697 	if (err < 0)
1698 		goto tap_ops_failed;
1699 
1700 	err = rtnl_link_register(&erspan_link_ops);
1701 	if (err < 0)
1702 		goto erspan_link_failed;
1703 
1704 	return 0;
1705 
1706 erspan_link_failed:
1707 	rtnl_link_unregister(&ipgre_tap_ops);
1708 tap_ops_failed:
1709 	rtnl_link_unregister(&ipgre_link_ops);
1710 rtnl_link_failed:
1711 	gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1712 add_proto_failed:
1713 	unregister_pernet_device(&erspan_net_ops);
1714 pnet_erspan_failed:
1715 	unregister_pernet_device(&ipgre_tap_net_ops);
1716 pnet_tap_failed:
1717 	unregister_pernet_device(&ipgre_net_ops);
1718 	return err;
1719 }
1720 
1721 static void __exit ipgre_fini(void)
1722 {
1723 	rtnl_link_unregister(&ipgre_tap_ops);
1724 	rtnl_link_unregister(&ipgre_link_ops);
1725 	rtnl_link_unregister(&erspan_link_ops);
1726 	gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1727 	unregister_pernet_device(&ipgre_tap_net_ops);
1728 	unregister_pernet_device(&ipgre_net_ops);
1729 	unregister_pernet_device(&erspan_net_ops);
1730 }
1731 
1732 module_init(ipgre_init);
1733 module_exit(ipgre_fini);
1734 MODULE_LICENSE("GPL");
1735 MODULE_ALIAS_RTNL_LINK("gre");
1736 MODULE_ALIAS_RTNL_LINK("gretap");
1737 MODULE_ALIAS_RTNL_LINK("erspan");
1738 MODULE_ALIAS_NETDEV("gre0");
1739 MODULE_ALIAS_NETDEV("gretap0");
1740 MODULE_ALIAS_NETDEV("erspan0");
1741