xref: /linux/net/ipv4/ip_gre.c (revision 55f3538c4923e9dfca132e99ebec370e8094afda)
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 	int ver;
265 	int len;
266 
267 	itn = net_generic(net, erspan_net_id);
268 	len = gre_hdr_len + sizeof(*ershdr);
269 
270 	/* Check based hdr len */
271 	if (unlikely(!pskb_may_pull(skb, len)))
272 		return PACKET_REJECT;
273 
274 	iph = ip_hdr(skb);
275 	ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
276 	ver = ershdr->ver;
277 
278 	/* The original GRE header does not have key field,
279 	 * Use ERSPAN 10-bit session ID as key.
280 	 */
281 	tpi->key = cpu_to_be32(get_session_id(ershdr));
282 	tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex,
283 				  tpi->flags | TUNNEL_KEY,
284 				  iph->saddr, iph->daddr, tpi->key);
285 
286 	if (tunnel) {
287 		len = gre_hdr_len + erspan_hdr_len(ver);
288 		if (unlikely(!pskb_may_pull(skb, len)))
289 			return PACKET_REJECT;
290 
291 		ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
292 		pkt_md = (struct erspan_metadata *)(ershdr + 1);
293 
294 		if (__iptunnel_pull_header(skb,
295 					   len,
296 					   htons(ETH_P_TEB),
297 					   false, false) < 0)
298 			goto drop;
299 
300 		if (tunnel->collect_md) {
301 			struct ip_tunnel_info *info;
302 			struct erspan_metadata *md;
303 			__be64 tun_id;
304 			__be16 flags;
305 
306 			tpi->flags |= TUNNEL_KEY;
307 			flags = tpi->flags;
308 			tun_id = key32_to_tunnel_id(tpi->key);
309 
310 			tun_dst = ip_tun_rx_dst(skb, flags,
311 						tun_id, sizeof(*md));
312 			if (!tun_dst)
313 				return PACKET_REJECT;
314 
315 			md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
316 			memcpy(md, pkt_md, sizeof(*md));
317 			md->version = ver;
318 
319 			info = &tun_dst->u.tun_info;
320 			info->key.tun_flags |= TUNNEL_ERSPAN_OPT;
321 			info->options_len = sizeof(*md);
322 		} else {
323 			tunnel->erspan_ver = ver;
324 			if (ver == 1) {
325 				tunnel->index = ntohl(pkt_md->u.index);
326 			} else {
327 				tunnel->dir = pkt_md->u.md2.dir;
328 				tunnel->hwid = get_hwid(&pkt_md->u.md2);
329 			}
330 
331 		}
332 
333 		skb_reset_mac_header(skb);
334 		ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
335 		return PACKET_RCVD;
336 	}
337 drop:
338 	kfree_skb(skb);
339 	return PACKET_RCVD;
340 }
341 
342 static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
343 		       struct ip_tunnel_net *itn, int hdr_len, bool raw_proto)
344 {
345 	struct metadata_dst *tun_dst = NULL;
346 	const struct iphdr *iph;
347 	struct ip_tunnel *tunnel;
348 
349 	iph = ip_hdr(skb);
350 	tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
351 				  iph->saddr, iph->daddr, tpi->key);
352 
353 	if (tunnel) {
354 		if (__iptunnel_pull_header(skb, hdr_len, tpi->proto,
355 					   raw_proto, false) < 0)
356 			goto drop;
357 
358 		if (tunnel->dev->type != ARPHRD_NONE)
359 			skb_pop_mac_header(skb);
360 		else
361 			skb_reset_mac_header(skb);
362 		if (tunnel->collect_md) {
363 			__be16 flags;
364 			__be64 tun_id;
365 
366 			flags = tpi->flags & (TUNNEL_CSUM | TUNNEL_KEY);
367 			tun_id = key32_to_tunnel_id(tpi->key);
368 			tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
369 			if (!tun_dst)
370 				return PACKET_REJECT;
371 		}
372 
373 		ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
374 		return PACKET_RCVD;
375 	}
376 	return PACKET_NEXT;
377 
378 drop:
379 	kfree_skb(skb);
380 	return PACKET_RCVD;
381 }
382 
383 static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
384 		     int hdr_len)
385 {
386 	struct net *net = dev_net(skb->dev);
387 	struct ip_tunnel_net *itn;
388 	int res;
389 
390 	if (tpi->proto == htons(ETH_P_TEB))
391 		itn = net_generic(net, gre_tap_net_id);
392 	else
393 		itn = net_generic(net, ipgre_net_id);
394 
395 	res = __ipgre_rcv(skb, tpi, itn, hdr_len, false);
396 	if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) {
397 		/* ipgre tunnels in collect metadata mode should receive
398 		 * also ETH_P_TEB traffic.
399 		 */
400 		itn = net_generic(net, ipgre_net_id);
401 		res = __ipgre_rcv(skb, tpi, itn, hdr_len, true);
402 	}
403 	return res;
404 }
405 
406 static int gre_rcv(struct sk_buff *skb)
407 {
408 	struct tnl_ptk_info tpi;
409 	bool csum_err = false;
410 	int hdr_len;
411 
412 #ifdef CONFIG_NET_IPGRE_BROADCAST
413 	if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
414 		/* Looped back packet, drop it! */
415 		if (rt_is_output_route(skb_rtable(skb)))
416 			goto drop;
417 	}
418 #endif
419 
420 	hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0);
421 	if (hdr_len < 0)
422 		goto drop;
423 
424 	if (unlikely(tpi.proto == htons(ETH_P_ERSPAN) ||
425 		     tpi.proto == htons(ETH_P_ERSPAN2))) {
426 		if (erspan_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
427 			return 0;
428 		goto out;
429 	}
430 
431 	if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
432 		return 0;
433 
434 out:
435 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
436 drop:
437 	kfree_skb(skb);
438 	return 0;
439 }
440 
441 static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
442 		       const struct iphdr *tnl_params,
443 		       __be16 proto)
444 {
445 	struct ip_tunnel *tunnel = netdev_priv(dev);
446 
447 	if (tunnel->parms.o_flags & TUNNEL_SEQ)
448 		tunnel->o_seqno++;
449 
450 	/* Push GRE header. */
451 	gre_build_header(skb, tunnel->tun_hlen,
452 			 tunnel->parms.o_flags, proto, tunnel->parms.o_key,
453 			 htonl(tunnel->o_seqno));
454 
455 	ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
456 }
457 
458 static int gre_handle_offloads(struct sk_buff *skb, bool csum)
459 {
460 	return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
461 }
462 
463 static struct rtable *gre_get_rt(struct sk_buff *skb,
464 				 struct net_device *dev,
465 				 struct flowi4 *fl,
466 				 const struct ip_tunnel_key *key)
467 {
468 	struct net *net = dev_net(dev);
469 
470 	memset(fl, 0, sizeof(*fl));
471 	fl->daddr = key->u.ipv4.dst;
472 	fl->saddr = key->u.ipv4.src;
473 	fl->flowi4_tos = RT_TOS(key->tos);
474 	fl->flowi4_mark = skb->mark;
475 	fl->flowi4_proto = IPPROTO_GRE;
476 
477 	return ip_route_output_key(net, fl);
478 }
479 
480 static struct rtable *prepare_fb_xmit(struct sk_buff *skb,
481 				      struct net_device *dev,
482 				      struct flowi4 *fl,
483 				      int tunnel_hlen)
484 {
485 	struct ip_tunnel_info *tun_info;
486 	const struct ip_tunnel_key *key;
487 	struct rtable *rt = NULL;
488 	int min_headroom;
489 	bool use_cache;
490 	int err;
491 
492 	tun_info = skb_tunnel_info(skb);
493 	key = &tun_info->key;
494 	use_cache = ip_tunnel_dst_cache_usable(skb, tun_info);
495 
496 	if (use_cache)
497 		rt = dst_cache_get_ip4(&tun_info->dst_cache, &fl->saddr);
498 	if (!rt) {
499 		rt = gre_get_rt(skb, dev, fl, key);
500 		if (IS_ERR(rt))
501 			goto err_free_skb;
502 		if (use_cache)
503 			dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst,
504 					  fl->saddr);
505 	}
506 
507 	min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
508 			+ tunnel_hlen + sizeof(struct iphdr);
509 	if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
510 		int head_delta = SKB_DATA_ALIGN(min_headroom -
511 						skb_headroom(skb) +
512 						16);
513 		err = pskb_expand_head(skb, max_t(int, head_delta, 0),
514 				       0, GFP_ATOMIC);
515 		if (unlikely(err))
516 			goto err_free_rt;
517 	}
518 	return rt;
519 
520 err_free_rt:
521 	ip_rt_put(rt);
522 err_free_skb:
523 	kfree_skb(skb);
524 	dev->stats.tx_dropped++;
525 	return NULL;
526 }
527 
528 static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev,
529 			__be16 proto)
530 {
531 	struct ip_tunnel_info *tun_info;
532 	const struct ip_tunnel_key *key;
533 	struct rtable *rt = NULL;
534 	struct flowi4 fl;
535 	int tunnel_hlen;
536 	__be16 df, flags;
537 
538 	tun_info = skb_tunnel_info(skb);
539 	if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
540 		     ip_tunnel_info_af(tun_info) != AF_INET))
541 		goto err_free_skb;
542 
543 	key = &tun_info->key;
544 	tunnel_hlen = gre_calc_hlen(key->tun_flags);
545 
546 	rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
547 	if (!rt)
548 		return;
549 
550 	/* Push Tunnel header. */
551 	if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM)))
552 		goto err_free_rt;
553 
554 	flags = tun_info->key.tun_flags & (TUNNEL_CSUM | TUNNEL_KEY);
555 	gre_build_header(skb, tunnel_hlen, flags, proto,
556 			 tunnel_id_to_key32(tun_info->key.tun_id), 0);
557 
558 	df = key->tun_flags & TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
559 
560 	iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
561 		      key->tos, key->ttl, df, false);
562 	return;
563 
564 err_free_rt:
565 	ip_rt_put(rt);
566 err_free_skb:
567 	kfree_skb(skb);
568 	dev->stats.tx_dropped++;
569 }
570 
571 static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev,
572 			   __be16 proto)
573 {
574 	struct ip_tunnel *tunnel = netdev_priv(dev);
575 	struct ip_tunnel_info *tun_info;
576 	const struct ip_tunnel_key *key;
577 	struct erspan_metadata *md;
578 	struct rtable *rt = NULL;
579 	bool truncate = false;
580 	struct flowi4 fl;
581 	int tunnel_hlen;
582 	int version;
583 	__be16 df;
584 
585 	tun_info = skb_tunnel_info(skb);
586 	if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
587 		     ip_tunnel_info_af(tun_info) != AF_INET))
588 		goto err_free_skb;
589 
590 	key = &tun_info->key;
591 	md = ip_tunnel_info_opts(tun_info);
592 	if (!md)
593 		goto err_free_rt;
594 
595 	/* ERSPAN has fixed 8 byte GRE header */
596 	version = md->version;
597 	tunnel_hlen = 8 + erspan_hdr_len(version);
598 
599 	rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
600 	if (!rt)
601 		return;
602 
603 	if (gre_handle_offloads(skb, false))
604 		goto err_free_rt;
605 
606 	if (skb->len > dev->mtu + dev->hard_header_len) {
607 		pskb_trim(skb, dev->mtu + dev->hard_header_len);
608 		truncate = true;
609 	}
610 
611 	if (version == 1) {
612 		erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)),
613 				    ntohl(md->u.index), truncate, true);
614 	} else if (version == 2) {
615 		erspan_build_header_v2(skb,
616 				       ntohl(tunnel_id_to_key32(key->tun_id)),
617 				       md->u.md2.dir,
618 				       get_hwid(&md->u.md2),
619 				       truncate, true);
620 	} else {
621 		goto err_free_rt;
622 	}
623 
624 	gre_build_header(skb, 8, TUNNEL_SEQ,
625 			 htons(ETH_P_ERSPAN), 0, htonl(tunnel->o_seqno++));
626 
627 	df = key->tun_flags & TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
628 
629 	iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
630 		      key->tos, key->ttl, df, false);
631 	return;
632 
633 err_free_rt:
634 	ip_rt_put(rt);
635 err_free_skb:
636 	kfree_skb(skb);
637 	dev->stats.tx_dropped++;
638 }
639 
640 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
641 {
642 	struct ip_tunnel_info *info = skb_tunnel_info(skb);
643 	struct rtable *rt;
644 	struct flowi4 fl4;
645 
646 	if (ip_tunnel_info_af(info) != AF_INET)
647 		return -EINVAL;
648 
649 	rt = gre_get_rt(skb, dev, &fl4, &info->key);
650 	if (IS_ERR(rt))
651 		return PTR_ERR(rt);
652 
653 	ip_rt_put(rt);
654 	info->key.u.ipv4.src = fl4.saddr;
655 	return 0;
656 }
657 
658 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
659 			      struct net_device *dev)
660 {
661 	struct ip_tunnel *tunnel = netdev_priv(dev);
662 	const struct iphdr *tnl_params;
663 
664 	if (tunnel->collect_md) {
665 		gre_fb_xmit(skb, dev, skb->protocol);
666 		return NETDEV_TX_OK;
667 	}
668 
669 	if (dev->header_ops) {
670 		/* Need space for new headers */
671 		if (skb_cow_head(skb, dev->needed_headroom -
672 				      (tunnel->hlen + sizeof(struct iphdr))))
673 			goto free_skb;
674 
675 		tnl_params = (const struct iphdr *)skb->data;
676 
677 		/* Pull skb since ip_tunnel_xmit() needs skb->data pointing
678 		 * to gre header.
679 		 */
680 		skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
681 		skb_reset_mac_header(skb);
682 	} else {
683 		if (skb_cow_head(skb, dev->needed_headroom))
684 			goto free_skb;
685 
686 		tnl_params = &tunnel->parms.iph;
687 	}
688 
689 	if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
690 		goto free_skb;
691 
692 	__gre_xmit(skb, dev, tnl_params, skb->protocol);
693 	return NETDEV_TX_OK;
694 
695 free_skb:
696 	kfree_skb(skb);
697 	dev->stats.tx_dropped++;
698 	return NETDEV_TX_OK;
699 }
700 
701 static netdev_tx_t erspan_xmit(struct sk_buff *skb,
702 			       struct net_device *dev)
703 {
704 	struct ip_tunnel *tunnel = netdev_priv(dev);
705 	bool truncate = false;
706 
707 	if (tunnel->collect_md) {
708 		erspan_fb_xmit(skb, dev, skb->protocol);
709 		return NETDEV_TX_OK;
710 	}
711 
712 	if (gre_handle_offloads(skb, false))
713 		goto free_skb;
714 
715 	if (skb_cow_head(skb, dev->needed_headroom))
716 		goto free_skb;
717 
718 	if (skb->len > dev->mtu + dev->hard_header_len) {
719 		pskb_trim(skb, dev->mtu + dev->hard_header_len);
720 		truncate = true;
721 	}
722 
723 	/* Push ERSPAN header */
724 	if (tunnel->erspan_ver == 1)
725 		erspan_build_header(skb, ntohl(tunnel->parms.o_key),
726 				    tunnel->index,
727 				    truncate, true);
728 	else
729 		erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key),
730 				       tunnel->dir, tunnel->hwid,
731 				       truncate, true);
732 
733 	tunnel->parms.o_flags &= ~TUNNEL_KEY;
734 	__gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_ERSPAN));
735 	return NETDEV_TX_OK;
736 
737 free_skb:
738 	kfree_skb(skb);
739 	dev->stats.tx_dropped++;
740 	return NETDEV_TX_OK;
741 }
742 
743 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
744 				struct net_device *dev)
745 {
746 	struct ip_tunnel *tunnel = netdev_priv(dev);
747 
748 	if (tunnel->collect_md) {
749 		gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
750 		return NETDEV_TX_OK;
751 	}
752 
753 	if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
754 		goto free_skb;
755 
756 	if (skb_cow_head(skb, dev->needed_headroom))
757 		goto free_skb;
758 
759 	__gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
760 	return NETDEV_TX_OK;
761 
762 free_skb:
763 	kfree_skb(skb);
764 	dev->stats.tx_dropped++;
765 	return NETDEV_TX_OK;
766 }
767 
768 static void ipgre_link_update(struct net_device *dev, bool set_mtu)
769 {
770 	struct ip_tunnel *tunnel = netdev_priv(dev);
771 	int len;
772 
773 	len = tunnel->tun_hlen;
774 	tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
775 	len = tunnel->tun_hlen - len;
776 	tunnel->hlen = tunnel->hlen + len;
777 
778 	dev->needed_headroom = dev->needed_headroom + len;
779 	if (set_mtu)
780 		dev->mtu = max_t(int, dev->mtu - len, 68);
781 
782 	if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
783 		if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
784 		    tunnel->encap.type == TUNNEL_ENCAP_NONE) {
785 			dev->features |= NETIF_F_GSO_SOFTWARE;
786 			dev->hw_features |= NETIF_F_GSO_SOFTWARE;
787 		}
788 		dev->features |= NETIF_F_LLTX;
789 	}
790 }
791 
792 static int ipgre_tunnel_ioctl(struct net_device *dev,
793 			      struct ifreq *ifr, int cmd)
794 {
795 	struct ip_tunnel_parm p;
796 	int err;
797 
798 	if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
799 		return -EFAULT;
800 
801 	if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
802 		if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
803 		    p.iph.ihl != 5 || (p.iph.frag_off & htons(~IP_DF)) ||
804 		    ((p.i_flags | p.o_flags) & (GRE_VERSION | GRE_ROUTING)))
805 			return -EINVAL;
806 	}
807 
808 	p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
809 	p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
810 
811 	err = ip_tunnel_ioctl(dev, &p, cmd);
812 	if (err)
813 		return err;
814 
815 	if (cmd == SIOCCHGTUNNEL) {
816 		struct ip_tunnel *t = netdev_priv(dev);
817 
818 		t->parms.i_flags = p.i_flags;
819 		t->parms.o_flags = p.o_flags;
820 
821 		if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
822 			ipgre_link_update(dev, true);
823 	}
824 
825 	p.i_flags = gre_tnl_flags_to_gre_flags(p.i_flags);
826 	p.o_flags = gre_tnl_flags_to_gre_flags(p.o_flags);
827 
828 	if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
829 		return -EFAULT;
830 
831 	return 0;
832 }
833 
834 /* Nice toy. Unfortunately, useless in real life :-)
835    It allows to construct virtual multiprotocol broadcast "LAN"
836    over the Internet, provided multicast routing is tuned.
837 
838 
839    I have no idea was this bicycle invented before me,
840    so that I had to set ARPHRD_IPGRE to a random value.
841    I have an impression, that Cisco could make something similar,
842    but this feature is apparently missing in IOS<=11.2(8).
843 
844    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
845    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
846 
847    ping -t 255 224.66.66.66
848 
849    If nobody answers, mbone does not work.
850 
851    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
852    ip addr add 10.66.66.<somewhat>/24 dev Universe
853    ifconfig Universe up
854    ifconfig Universe add fe80::<Your_real_addr>/10
855    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
856    ftp 10.66.66.66
857    ...
858    ftp fec0:6666:6666::193.233.7.65
859    ...
860  */
861 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
862 			unsigned short type,
863 			const void *daddr, const void *saddr, unsigned int len)
864 {
865 	struct ip_tunnel *t = netdev_priv(dev);
866 	struct iphdr *iph;
867 	struct gre_base_hdr *greh;
868 
869 	iph = skb_push(skb, t->hlen + sizeof(*iph));
870 	greh = (struct gre_base_hdr *)(iph+1);
871 	greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
872 	greh->protocol = htons(type);
873 
874 	memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
875 
876 	/* Set the source hardware address. */
877 	if (saddr)
878 		memcpy(&iph->saddr, saddr, 4);
879 	if (daddr)
880 		memcpy(&iph->daddr, daddr, 4);
881 	if (iph->daddr)
882 		return t->hlen + sizeof(*iph);
883 
884 	return -(t->hlen + sizeof(*iph));
885 }
886 
887 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
888 {
889 	const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
890 	memcpy(haddr, &iph->saddr, 4);
891 	return 4;
892 }
893 
894 static const struct header_ops ipgre_header_ops = {
895 	.create	= ipgre_header,
896 	.parse	= ipgre_header_parse,
897 };
898 
899 #ifdef CONFIG_NET_IPGRE_BROADCAST
900 static int ipgre_open(struct net_device *dev)
901 {
902 	struct ip_tunnel *t = netdev_priv(dev);
903 
904 	if (ipv4_is_multicast(t->parms.iph.daddr)) {
905 		struct flowi4 fl4;
906 		struct rtable *rt;
907 
908 		rt = ip_route_output_gre(t->net, &fl4,
909 					 t->parms.iph.daddr,
910 					 t->parms.iph.saddr,
911 					 t->parms.o_key,
912 					 RT_TOS(t->parms.iph.tos),
913 					 t->parms.link);
914 		if (IS_ERR(rt))
915 			return -EADDRNOTAVAIL;
916 		dev = rt->dst.dev;
917 		ip_rt_put(rt);
918 		if (!__in_dev_get_rtnl(dev))
919 			return -EADDRNOTAVAIL;
920 		t->mlink = dev->ifindex;
921 		ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
922 	}
923 	return 0;
924 }
925 
926 static int ipgre_close(struct net_device *dev)
927 {
928 	struct ip_tunnel *t = netdev_priv(dev);
929 
930 	if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
931 		struct in_device *in_dev;
932 		in_dev = inetdev_by_index(t->net, t->mlink);
933 		if (in_dev)
934 			ip_mc_dec_group(in_dev, t->parms.iph.daddr);
935 	}
936 	return 0;
937 }
938 #endif
939 
940 static const struct net_device_ops ipgre_netdev_ops = {
941 	.ndo_init		= ipgre_tunnel_init,
942 	.ndo_uninit		= ip_tunnel_uninit,
943 #ifdef CONFIG_NET_IPGRE_BROADCAST
944 	.ndo_open		= ipgre_open,
945 	.ndo_stop		= ipgre_close,
946 #endif
947 	.ndo_start_xmit		= ipgre_xmit,
948 	.ndo_do_ioctl		= ipgre_tunnel_ioctl,
949 	.ndo_change_mtu		= ip_tunnel_change_mtu,
950 	.ndo_get_stats64	= ip_tunnel_get_stats64,
951 	.ndo_get_iflink		= ip_tunnel_get_iflink,
952 };
953 
954 #define GRE_FEATURES (NETIF_F_SG |		\
955 		      NETIF_F_FRAGLIST |	\
956 		      NETIF_F_HIGHDMA |		\
957 		      NETIF_F_HW_CSUM)
958 
959 static void ipgre_tunnel_setup(struct net_device *dev)
960 {
961 	dev->netdev_ops		= &ipgre_netdev_ops;
962 	dev->type		= ARPHRD_IPGRE;
963 	ip_tunnel_setup(dev, ipgre_net_id);
964 }
965 
966 static void __gre_tunnel_init(struct net_device *dev)
967 {
968 	struct ip_tunnel *tunnel;
969 	int t_hlen;
970 
971 	tunnel = netdev_priv(dev);
972 	tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
973 	tunnel->parms.iph.protocol = IPPROTO_GRE;
974 
975 	tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
976 
977 	t_hlen = tunnel->hlen + sizeof(struct iphdr);
978 
979 	dev->needed_headroom	= LL_MAX_HEADER + t_hlen + 4;
980 	dev->mtu		= ETH_DATA_LEN - t_hlen - 4;
981 
982 	dev->features		|= GRE_FEATURES;
983 	dev->hw_features	|= GRE_FEATURES;
984 
985 	if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
986 		/* TCP offload with GRE SEQ is not supported, nor
987 		 * can we support 2 levels of outer headers requiring
988 		 * an update.
989 		 */
990 		if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
991 		    (tunnel->encap.type == TUNNEL_ENCAP_NONE)) {
992 			dev->features    |= NETIF_F_GSO_SOFTWARE;
993 			dev->hw_features |= NETIF_F_GSO_SOFTWARE;
994 		}
995 
996 		/* Can use a lockless transmit, unless we generate
997 		 * output sequences
998 		 */
999 		dev->features |= NETIF_F_LLTX;
1000 	}
1001 }
1002 
1003 static int ipgre_tunnel_init(struct net_device *dev)
1004 {
1005 	struct ip_tunnel *tunnel = netdev_priv(dev);
1006 	struct iphdr *iph = &tunnel->parms.iph;
1007 
1008 	__gre_tunnel_init(dev);
1009 
1010 	memcpy(dev->dev_addr, &iph->saddr, 4);
1011 	memcpy(dev->broadcast, &iph->daddr, 4);
1012 
1013 	dev->flags		= IFF_NOARP;
1014 	netif_keep_dst(dev);
1015 	dev->addr_len		= 4;
1016 
1017 	if (iph->daddr && !tunnel->collect_md) {
1018 #ifdef CONFIG_NET_IPGRE_BROADCAST
1019 		if (ipv4_is_multicast(iph->daddr)) {
1020 			if (!iph->saddr)
1021 				return -EINVAL;
1022 			dev->flags = IFF_BROADCAST;
1023 			dev->header_ops = &ipgre_header_ops;
1024 		}
1025 #endif
1026 	} else if (!tunnel->collect_md) {
1027 		dev->header_ops = &ipgre_header_ops;
1028 	}
1029 
1030 	return ip_tunnel_init(dev);
1031 }
1032 
1033 static const struct gre_protocol ipgre_protocol = {
1034 	.handler     = gre_rcv,
1035 	.err_handler = gre_err,
1036 };
1037 
1038 static int __net_init ipgre_init_net(struct net *net)
1039 {
1040 	return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
1041 }
1042 
1043 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net)
1044 {
1045 	ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops);
1046 }
1047 
1048 static struct pernet_operations ipgre_net_ops = {
1049 	.init = ipgre_init_net,
1050 	.exit_batch = ipgre_exit_batch_net,
1051 	.id   = &ipgre_net_id,
1052 	.size = sizeof(struct ip_tunnel_net),
1053 };
1054 
1055 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1056 				 struct netlink_ext_ack *extack)
1057 {
1058 	__be16 flags;
1059 
1060 	if (!data)
1061 		return 0;
1062 
1063 	flags = 0;
1064 	if (data[IFLA_GRE_IFLAGS])
1065 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1066 	if (data[IFLA_GRE_OFLAGS])
1067 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1068 	if (flags & (GRE_VERSION|GRE_ROUTING))
1069 		return -EINVAL;
1070 
1071 	if (data[IFLA_GRE_COLLECT_METADATA] &&
1072 	    data[IFLA_GRE_ENCAP_TYPE] &&
1073 	    nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1074 		return -EINVAL;
1075 
1076 	return 0;
1077 }
1078 
1079 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1080 			      struct netlink_ext_ack *extack)
1081 {
1082 	__be32 daddr;
1083 
1084 	if (tb[IFLA_ADDRESS]) {
1085 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1086 			return -EINVAL;
1087 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1088 			return -EADDRNOTAVAIL;
1089 	}
1090 
1091 	if (!data)
1092 		goto out;
1093 
1094 	if (data[IFLA_GRE_REMOTE]) {
1095 		memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1096 		if (!daddr)
1097 			return -EINVAL;
1098 	}
1099 
1100 out:
1101 	return ipgre_tunnel_validate(tb, data, extack);
1102 }
1103 
1104 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1105 			   struct netlink_ext_ack *extack)
1106 {
1107 	__be16 flags = 0;
1108 	int ret;
1109 
1110 	if (!data)
1111 		return 0;
1112 
1113 	ret = ipgre_tap_validate(tb, data, extack);
1114 	if (ret)
1115 		return ret;
1116 
1117 	/* ERSPAN should only have GRE sequence and key flag */
1118 	if (data[IFLA_GRE_OFLAGS])
1119 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1120 	if (data[IFLA_GRE_IFLAGS])
1121 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1122 	if (!data[IFLA_GRE_COLLECT_METADATA] &&
1123 	    flags != (GRE_SEQ | GRE_KEY))
1124 		return -EINVAL;
1125 
1126 	/* ERSPAN Session ID only has 10-bit. Since we reuse
1127 	 * 32-bit key field as ID, check it's range.
1128 	 */
1129 	if (data[IFLA_GRE_IKEY] &&
1130 	    (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1131 		return -EINVAL;
1132 
1133 	if (data[IFLA_GRE_OKEY] &&
1134 	    (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1135 		return -EINVAL;
1136 
1137 	return 0;
1138 }
1139 
1140 static int ipgre_netlink_parms(struct net_device *dev,
1141 				struct nlattr *data[],
1142 				struct nlattr *tb[],
1143 				struct ip_tunnel_parm *parms,
1144 				__u32 *fwmark)
1145 {
1146 	struct ip_tunnel *t = netdev_priv(dev);
1147 
1148 	memset(parms, 0, sizeof(*parms));
1149 
1150 	parms->iph.protocol = IPPROTO_GRE;
1151 
1152 	if (!data)
1153 		return 0;
1154 
1155 	if (data[IFLA_GRE_LINK])
1156 		parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1157 
1158 	if (data[IFLA_GRE_IFLAGS])
1159 		parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
1160 
1161 	if (data[IFLA_GRE_OFLAGS])
1162 		parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
1163 
1164 	if (data[IFLA_GRE_IKEY])
1165 		parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1166 
1167 	if (data[IFLA_GRE_OKEY])
1168 		parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1169 
1170 	if (data[IFLA_GRE_LOCAL])
1171 		parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1172 
1173 	if (data[IFLA_GRE_REMOTE])
1174 		parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1175 
1176 	if (data[IFLA_GRE_TTL])
1177 		parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1178 
1179 	if (data[IFLA_GRE_TOS])
1180 		parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1181 
1182 	if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1183 		if (t->ignore_df)
1184 			return -EINVAL;
1185 		parms->iph.frag_off = htons(IP_DF);
1186 	}
1187 
1188 	if (data[IFLA_GRE_COLLECT_METADATA]) {
1189 		t->collect_md = true;
1190 		if (dev->type == ARPHRD_IPGRE)
1191 			dev->type = ARPHRD_NONE;
1192 	}
1193 
1194 	if (data[IFLA_GRE_IGNORE_DF]) {
1195 		if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1196 		  && (parms->iph.frag_off & htons(IP_DF)))
1197 			return -EINVAL;
1198 		t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1199 	}
1200 
1201 	if (data[IFLA_GRE_FWMARK])
1202 		*fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1203 
1204 	if (data[IFLA_GRE_ERSPAN_VER]) {
1205 		t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]);
1206 
1207 		if (t->erspan_ver != 1 && t->erspan_ver != 2)
1208 			return -EINVAL;
1209 	}
1210 
1211 	if (t->erspan_ver == 1) {
1212 		if (data[IFLA_GRE_ERSPAN_INDEX]) {
1213 			t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1214 			if (t->index & ~INDEX_MASK)
1215 				return -EINVAL;
1216 		}
1217 	} else if (t->erspan_ver == 2) {
1218 		if (data[IFLA_GRE_ERSPAN_DIR]) {
1219 			t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]);
1220 			if (t->dir & ~(DIR_MASK >> DIR_OFFSET))
1221 				return -EINVAL;
1222 		}
1223 		if (data[IFLA_GRE_ERSPAN_HWID]) {
1224 			t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]);
1225 			if (t->hwid & ~(HWID_MASK >> HWID_OFFSET))
1226 				return -EINVAL;
1227 		}
1228 	}
1229 
1230 	return 0;
1231 }
1232 
1233 /* This function returns true when ENCAP attributes are present in the nl msg */
1234 static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1235 				      struct ip_tunnel_encap *ipencap)
1236 {
1237 	bool ret = false;
1238 
1239 	memset(ipencap, 0, sizeof(*ipencap));
1240 
1241 	if (!data)
1242 		return ret;
1243 
1244 	if (data[IFLA_GRE_ENCAP_TYPE]) {
1245 		ret = true;
1246 		ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1247 	}
1248 
1249 	if (data[IFLA_GRE_ENCAP_FLAGS]) {
1250 		ret = true;
1251 		ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1252 	}
1253 
1254 	if (data[IFLA_GRE_ENCAP_SPORT]) {
1255 		ret = true;
1256 		ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1257 	}
1258 
1259 	if (data[IFLA_GRE_ENCAP_DPORT]) {
1260 		ret = true;
1261 		ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1262 	}
1263 
1264 	return ret;
1265 }
1266 
1267 static int gre_tap_init(struct net_device *dev)
1268 {
1269 	__gre_tunnel_init(dev);
1270 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1271 	netif_keep_dst(dev);
1272 
1273 	return ip_tunnel_init(dev);
1274 }
1275 
1276 static const struct net_device_ops gre_tap_netdev_ops = {
1277 	.ndo_init		= gre_tap_init,
1278 	.ndo_uninit		= ip_tunnel_uninit,
1279 	.ndo_start_xmit		= gre_tap_xmit,
1280 	.ndo_set_mac_address 	= eth_mac_addr,
1281 	.ndo_validate_addr	= eth_validate_addr,
1282 	.ndo_change_mtu		= ip_tunnel_change_mtu,
1283 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1284 	.ndo_get_iflink		= ip_tunnel_get_iflink,
1285 	.ndo_fill_metadata_dst	= gre_fill_metadata_dst,
1286 };
1287 
1288 static int erspan_tunnel_init(struct net_device *dev)
1289 {
1290 	struct ip_tunnel *tunnel = netdev_priv(dev);
1291 	int t_hlen;
1292 
1293 	tunnel->tun_hlen = 8;
1294 	tunnel->parms.iph.protocol = IPPROTO_GRE;
1295 	tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1296 		       erspan_hdr_len(tunnel->erspan_ver);
1297 	t_hlen = tunnel->hlen + sizeof(struct iphdr);
1298 
1299 	dev->needed_headroom = LL_MAX_HEADER + t_hlen + 4;
1300 	dev->mtu = ETH_DATA_LEN - t_hlen - 4;
1301 	dev->features		|= GRE_FEATURES;
1302 	dev->hw_features	|= GRE_FEATURES;
1303 	dev->priv_flags		|= IFF_LIVE_ADDR_CHANGE;
1304 	netif_keep_dst(dev);
1305 
1306 	return ip_tunnel_init(dev);
1307 }
1308 
1309 static const struct net_device_ops erspan_netdev_ops = {
1310 	.ndo_init		= erspan_tunnel_init,
1311 	.ndo_uninit		= ip_tunnel_uninit,
1312 	.ndo_start_xmit		= erspan_xmit,
1313 	.ndo_set_mac_address	= eth_mac_addr,
1314 	.ndo_validate_addr	= eth_validate_addr,
1315 	.ndo_change_mtu		= ip_tunnel_change_mtu,
1316 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1317 	.ndo_get_iflink		= ip_tunnel_get_iflink,
1318 	.ndo_fill_metadata_dst	= gre_fill_metadata_dst,
1319 };
1320 
1321 static void ipgre_tap_setup(struct net_device *dev)
1322 {
1323 	ether_setup(dev);
1324 	dev->max_mtu = 0;
1325 	dev->netdev_ops	= &gre_tap_netdev_ops;
1326 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1327 	dev->priv_flags	|= IFF_LIVE_ADDR_CHANGE;
1328 	ip_tunnel_setup(dev, gre_tap_net_id);
1329 }
1330 
1331 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1332 			 struct nlattr *tb[], struct nlattr *data[],
1333 			 struct netlink_ext_ack *extack)
1334 {
1335 	struct ip_tunnel_parm p;
1336 	struct ip_tunnel_encap ipencap;
1337 	__u32 fwmark = 0;
1338 	int err;
1339 
1340 	if (ipgre_netlink_encap_parms(data, &ipencap)) {
1341 		struct ip_tunnel *t = netdev_priv(dev);
1342 		err = ip_tunnel_encap_setup(t, &ipencap);
1343 
1344 		if (err < 0)
1345 			return err;
1346 	}
1347 
1348 	err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1349 	if (err < 0)
1350 		return err;
1351 	return ip_tunnel_newlink(dev, tb, &p, fwmark);
1352 }
1353 
1354 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1355 			    struct nlattr *data[],
1356 			    struct netlink_ext_ack *extack)
1357 {
1358 	struct ip_tunnel *t = netdev_priv(dev);
1359 	struct ip_tunnel_encap ipencap;
1360 	__u32 fwmark = t->fwmark;
1361 	struct ip_tunnel_parm p;
1362 	int err;
1363 
1364 	if (ipgre_netlink_encap_parms(data, &ipencap)) {
1365 		err = ip_tunnel_encap_setup(t, &ipencap);
1366 
1367 		if (err < 0)
1368 			return err;
1369 	}
1370 
1371 	err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1372 	if (err < 0)
1373 		return err;
1374 
1375 	err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1376 	if (err < 0)
1377 		return err;
1378 
1379 	t->parms.i_flags = p.i_flags;
1380 	t->parms.o_flags = p.o_flags;
1381 
1382 	if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
1383 		ipgre_link_update(dev, !tb[IFLA_MTU]);
1384 
1385 	return 0;
1386 }
1387 
1388 static size_t ipgre_get_size(const struct net_device *dev)
1389 {
1390 	return
1391 		/* IFLA_GRE_LINK */
1392 		nla_total_size(4) +
1393 		/* IFLA_GRE_IFLAGS */
1394 		nla_total_size(2) +
1395 		/* IFLA_GRE_OFLAGS */
1396 		nla_total_size(2) +
1397 		/* IFLA_GRE_IKEY */
1398 		nla_total_size(4) +
1399 		/* IFLA_GRE_OKEY */
1400 		nla_total_size(4) +
1401 		/* IFLA_GRE_LOCAL */
1402 		nla_total_size(4) +
1403 		/* IFLA_GRE_REMOTE */
1404 		nla_total_size(4) +
1405 		/* IFLA_GRE_TTL */
1406 		nla_total_size(1) +
1407 		/* IFLA_GRE_TOS */
1408 		nla_total_size(1) +
1409 		/* IFLA_GRE_PMTUDISC */
1410 		nla_total_size(1) +
1411 		/* IFLA_GRE_ENCAP_TYPE */
1412 		nla_total_size(2) +
1413 		/* IFLA_GRE_ENCAP_FLAGS */
1414 		nla_total_size(2) +
1415 		/* IFLA_GRE_ENCAP_SPORT */
1416 		nla_total_size(2) +
1417 		/* IFLA_GRE_ENCAP_DPORT */
1418 		nla_total_size(2) +
1419 		/* IFLA_GRE_COLLECT_METADATA */
1420 		nla_total_size(0) +
1421 		/* IFLA_GRE_IGNORE_DF */
1422 		nla_total_size(1) +
1423 		/* IFLA_GRE_FWMARK */
1424 		nla_total_size(4) +
1425 		/* IFLA_GRE_ERSPAN_INDEX */
1426 		nla_total_size(4) +
1427 		/* IFLA_GRE_ERSPAN_VER */
1428 		nla_total_size(1) +
1429 		/* IFLA_GRE_ERSPAN_DIR */
1430 		nla_total_size(1) +
1431 		/* IFLA_GRE_ERSPAN_HWID */
1432 		nla_total_size(2) +
1433 		0;
1434 }
1435 
1436 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1437 {
1438 	struct ip_tunnel *t = netdev_priv(dev);
1439 	struct ip_tunnel_parm *p = &t->parms;
1440 
1441 	if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1442 	    nla_put_be16(skb, IFLA_GRE_IFLAGS,
1443 			 gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1444 	    nla_put_be16(skb, IFLA_GRE_OFLAGS,
1445 			 gre_tnl_flags_to_gre_flags(p->o_flags)) ||
1446 	    nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1447 	    nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1448 	    nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1449 	    nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1450 	    nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1451 	    nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1452 	    nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1453 		       !!(p->iph.frag_off & htons(IP_DF))) ||
1454 	    nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1455 		goto nla_put_failure;
1456 
1457 	if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1458 			t->encap.type) ||
1459 	    nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1460 			 t->encap.sport) ||
1461 	    nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1462 			 t->encap.dport) ||
1463 	    nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1464 			t->encap.flags))
1465 		goto nla_put_failure;
1466 
1467 	if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1468 		goto nla_put_failure;
1469 
1470 	if (t->collect_md) {
1471 		if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1472 			goto nla_put_failure;
1473 	}
1474 
1475 	if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1476 		goto nla_put_failure;
1477 
1478 	if (t->erspan_ver == 1) {
1479 		if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1480 			goto nla_put_failure;
1481 	} else if (t->erspan_ver == 2) {
1482 		if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1483 			goto nla_put_failure;
1484 		if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1485 			goto nla_put_failure;
1486 	}
1487 
1488 	return 0;
1489 
1490 nla_put_failure:
1491 	return -EMSGSIZE;
1492 }
1493 
1494 static void erspan_setup(struct net_device *dev)
1495 {
1496 	ether_setup(dev);
1497 	dev->netdev_ops = &erspan_netdev_ops;
1498 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1499 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1500 	ip_tunnel_setup(dev, erspan_net_id);
1501 }
1502 
1503 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1504 	[IFLA_GRE_LINK]		= { .type = NLA_U32 },
1505 	[IFLA_GRE_IFLAGS]	= { .type = NLA_U16 },
1506 	[IFLA_GRE_OFLAGS]	= { .type = NLA_U16 },
1507 	[IFLA_GRE_IKEY]		= { .type = NLA_U32 },
1508 	[IFLA_GRE_OKEY]		= { .type = NLA_U32 },
1509 	[IFLA_GRE_LOCAL]	= { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1510 	[IFLA_GRE_REMOTE]	= { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1511 	[IFLA_GRE_TTL]		= { .type = NLA_U8 },
1512 	[IFLA_GRE_TOS]		= { .type = NLA_U8 },
1513 	[IFLA_GRE_PMTUDISC]	= { .type = NLA_U8 },
1514 	[IFLA_GRE_ENCAP_TYPE]	= { .type = NLA_U16 },
1515 	[IFLA_GRE_ENCAP_FLAGS]	= { .type = NLA_U16 },
1516 	[IFLA_GRE_ENCAP_SPORT]	= { .type = NLA_U16 },
1517 	[IFLA_GRE_ENCAP_DPORT]	= { .type = NLA_U16 },
1518 	[IFLA_GRE_COLLECT_METADATA]	= { .type = NLA_FLAG },
1519 	[IFLA_GRE_IGNORE_DF]	= { .type = NLA_U8 },
1520 	[IFLA_GRE_FWMARK]	= { .type = NLA_U32 },
1521 	[IFLA_GRE_ERSPAN_INDEX]	= { .type = NLA_U32 },
1522 	[IFLA_GRE_ERSPAN_VER]	= { .type = NLA_U8 },
1523 	[IFLA_GRE_ERSPAN_DIR]	= { .type = NLA_U8 },
1524 	[IFLA_GRE_ERSPAN_HWID]	= { .type = NLA_U16 },
1525 };
1526 
1527 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1528 	.kind		= "gre",
1529 	.maxtype	= IFLA_GRE_MAX,
1530 	.policy		= ipgre_policy,
1531 	.priv_size	= sizeof(struct ip_tunnel),
1532 	.setup		= ipgre_tunnel_setup,
1533 	.validate	= ipgre_tunnel_validate,
1534 	.newlink	= ipgre_newlink,
1535 	.changelink	= ipgre_changelink,
1536 	.dellink	= ip_tunnel_dellink,
1537 	.get_size	= ipgre_get_size,
1538 	.fill_info	= ipgre_fill_info,
1539 	.get_link_net	= ip_tunnel_get_link_net,
1540 };
1541 
1542 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1543 	.kind		= "gretap",
1544 	.maxtype	= IFLA_GRE_MAX,
1545 	.policy		= ipgre_policy,
1546 	.priv_size	= sizeof(struct ip_tunnel),
1547 	.setup		= ipgre_tap_setup,
1548 	.validate	= ipgre_tap_validate,
1549 	.newlink	= ipgre_newlink,
1550 	.changelink	= ipgre_changelink,
1551 	.dellink	= ip_tunnel_dellink,
1552 	.get_size	= ipgre_get_size,
1553 	.fill_info	= ipgre_fill_info,
1554 	.get_link_net	= ip_tunnel_get_link_net,
1555 };
1556 
1557 static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1558 	.kind		= "erspan",
1559 	.maxtype	= IFLA_GRE_MAX,
1560 	.policy		= ipgre_policy,
1561 	.priv_size	= sizeof(struct ip_tunnel),
1562 	.setup		= erspan_setup,
1563 	.validate	= erspan_validate,
1564 	.newlink	= ipgre_newlink,
1565 	.changelink	= ipgre_changelink,
1566 	.dellink	= ip_tunnel_dellink,
1567 	.get_size	= ipgre_get_size,
1568 	.fill_info	= ipgre_fill_info,
1569 	.get_link_net	= ip_tunnel_get_link_net,
1570 };
1571 
1572 struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1573 					u8 name_assign_type)
1574 {
1575 	struct nlattr *tb[IFLA_MAX + 1];
1576 	struct net_device *dev;
1577 	LIST_HEAD(list_kill);
1578 	struct ip_tunnel *t;
1579 	int err;
1580 
1581 	memset(&tb, 0, sizeof(tb));
1582 
1583 	dev = rtnl_create_link(net, name, name_assign_type,
1584 			       &ipgre_tap_ops, tb);
1585 	if (IS_ERR(dev))
1586 		return dev;
1587 
1588 	/* Configure flow based GRE device. */
1589 	t = netdev_priv(dev);
1590 	t->collect_md = true;
1591 
1592 	err = ipgre_newlink(net, dev, tb, NULL, NULL);
1593 	if (err < 0) {
1594 		free_netdev(dev);
1595 		return ERR_PTR(err);
1596 	}
1597 
1598 	/* openvswitch users expect packet sizes to be unrestricted,
1599 	 * so set the largest MTU we can.
1600 	 */
1601 	err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1602 	if (err)
1603 		goto out;
1604 
1605 	err = rtnl_configure_link(dev, NULL);
1606 	if (err < 0)
1607 		goto out;
1608 
1609 	return dev;
1610 out:
1611 	ip_tunnel_dellink(dev, &list_kill);
1612 	unregister_netdevice_many(&list_kill);
1613 	return ERR_PTR(err);
1614 }
1615 EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1616 
1617 static int __net_init ipgre_tap_init_net(struct net *net)
1618 {
1619 	return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1620 }
1621 
1622 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net)
1623 {
1624 	ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops);
1625 }
1626 
1627 static struct pernet_operations ipgre_tap_net_ops = {
1628 	.init = ipgre_tap_init_net,
1629 	.exit_batch = ipgre_tap_exit_batch_net,
1630 	.id   = &gre_tap_net_id,
1631 	.size = sizeof(struct ip_tunnel_net),
1632 };
1633 
1634 static int __net_init erspan_init_net(struct net *net)
1635 {
1636 	return ip_tunnel_init_net(net, erspan_net_id,
1637 				  &erspan_link_ops, "erspan0");
1638 }
1639 
1640 static void __net_exit erspan_exit_batch_net(struct list_head *net_list)
1641 {
1642 	ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops);
1643 }
1644 
1645 static struct pernet_operations erspan_net_ops = {
1646 	.init = erspan_init_net,
1647 	.exit_batch = erspan_exit_batch_net,
1648 	.id   = &erspan_net_id,
1649 	.size = sizeof(struct ip_tunnel_net),
1650 };
1651 
1652 static int __init ipgre_init(void)
1653 {
1654 	int err;
1655 
1656 	pr_info("GRE over IPv4 tunneling driver\n");
1657 
1658 	err = register_pernet_device(&ipgre_net_ops);
1659 	if (err < 0)
1660 		return err;
1661 
1662 	err = register_pernet_device(&ipgre_tap_net_ops);
1663 	if (err < 0)
1664 		goto pnet_tap_failed;
1665 
1666 	err = register_pernet_device(&erspan_net_ops);
1667 	if (err < 0)
1668 		goto pnet_erspan_failed;
1669 
1670 	err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1671 	if (err < 0) {
1672 		pr_info("%s: can't add protocol\n", __func__);
1673 		goto add_proto_failed;
1674 	}
1675 
1676 	err = rtnl_link_register(&ipgre_link_ops);
1677 	if (err < 0)
1678 		goto rtnl_link_failed;
1679 
1680 	err = rtnl_link_register(&ipgre_tap_ops);
1681 	if (err < 0)
1682 		goto tap_ops_failed;
1683 
1684 	err = rtnl_link_register(&erspan_link_ops);
1685 	if (err < 0)
1686 		goto erspan_link_failed;
1687 
1688 	return 0;
1689 
1690 erspan_link_failed:
1691 	rtnl_link_unregister(&ipgre_tap_ops);
1692 tap_ops_failed:
1693 	rtnl_link_unregister(&ipgre_link_ops);
1694 rtnl_link_failed:
1695 	gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1696 add_proto_failed:
1697 	unregister_pernet_device(&erspan_net_ops);
1698 pnet_erspan_failed:
1699 	unregister_pernet_device(&ipgre_tap_net_ops);
1700 pnet_tap_failed:
1701 	unregister_pernet_device(&ipgre_net_ops);
1702 	return err;
1703 }
1704 
1705 static void __exit ipgre_fini(void)
1706 {
1707 	rtnl_link_unregister(&ipgre_tap_ops);
1708 	rtnl_link_unregister(&ipgre_link_ops);
1709 	rtnl_link_unregister(&erspan_link_ops);
1710 	gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1711 	unregister_pernet_device(&ipgre_tap_net_ops);
1712 	unregister_pernet_device(&ipgre_net_ops);
1713 	unregister_pernet_device(&erspan_net_ops);
1714 }
1715 
1716 module_init(ipgre_init);
1717 module_exit(ipgre_fini);
1718 MODULE_LICENSE("GPL");
1719 MODULE_ALIAS_RTNL_LINK("gre");
1720 MODULE_ALIAS_RTNL_LINK("gretap");
1721 MODULE_ALIAS_RTNL_LINK("erspan");
1722 MODULE_ALIAS_NETDEV("gre0");
1723 MODULE_ALIAS_NETDEV("gretap0");
1724 MODULE_ALIAS_NETDEV("erspan0");
1725