xref: /openbmc/linux/net/ipv4/ip_gre.c (revision 4cff79e9)
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 
582 	tun_info = skb_tunnel_info(skb);
583 	if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
584 		     ip_tunnel_info_af(tun_info) != AF_INET))
585 		goto err_free_skb;
586 
587 	key = &tun_info->key;
588 	md = ip_tunnel_info_opts(tun_info);
589 	if (!md)
590 		goto err_free_rt;
591 
592 	/* ERSPAN has fixed 8 byte GRE header */
593 	version = md->version;
594 	tunnel_hlen = 8 + erspan_hdr_len(version);
595 
596 	rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
597 	if (!rt)
598 		return;
599 
600 	if (gre_handle_offloads(skb, false))
601 		goto err_free_rt;
602 
603 	if (skb->len > dev->mtu + dev->hard_header_len) {
604 		pskb_trim(skb, dev->mtu + dev->hard_header_len);
605 		truncate = true;
606 	}
607 
608 	if (version == 1) {
609 		erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)),
610 				    ntohl(md->u.index), truncate, true);
611 	} else if (version == 2) {
612 		erspan_build_header_v2(skb,
613 				       ntohl(tunnel_id_to_key32(key->tun_id)),
614 				       md->u.md2.dir,
615 				       get_hwid(&md->u.md2),
616 				       truncate, true);
617 	} else {
618 		goto err_free_rt;
619 	}
620 
621 	gre_build_header(skb, 8, TUNNEL_SEQ,
622 			 htons(ETH_P_ERSPAN), 0, htonl(tunnel->o_seqno++));
623 
624 	df = key->tun_flags & TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
625 
626 	iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
627 		      key->tos, key->ttl, df, false);
628 	return;
629 
630 err_free_rt:
631 	ip_rt_put(rt);
632 err_free_skb:
633 	kfree_skb(skb);
634 	dev->stats.tx_dropped++;
635 }
636 
637 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
638 {
639 	struct ip_tunnel_info *info = skb_tunnel_info(skb);
640 	struct rtable *rt;
641 	struct flowi4 fl4;
642 
643 	if (ip_tunnel_info_af(info) != AF_INET)
644 		return -EINVAL;
645 
646 	rt = gre_get_rt(skb, dev, &fl4, &info->key);
647 	if (IS_ERR(rt))
648 		return PTR_ERR(rt);
649 
650 	ip_rt_put(rt);
651 	info->key.u.ipv4.src = fl4.saddr;
652 	return 0;
653 }
654 
655 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
656 			      struct net_device *dev)
657 {
658 	struct ip_tunnel *tunnel = netdev_priv(dev);
659 	const struct iphdr *tnl_params;
660 
661 	if (tunnel->collect_md) {
662 		gre_fb_xmit(skb, dev, skb->protocol);
663 		return NETDEV_TX_OK;
664 	}
665 
666 	if (dev->header_ops) {
667 		/* Need space for new headers */
668 		if (skb_cow_head(skb, dev->needed_headroom -
669 				      (tunnel->hlen + sizeof(struct iphdr))))
670 			goto free_skb;
671 
672 		tnl_params = (const struct iphdr *)skb->data;
673 
674 		/* Pull skb since ip_tunnel_xmit() needs skb->data pointing
675 		 * to gre header.
676 		 */
677 		skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
678 		skb_reset_mac_header(skb);
679 	} else {
680 		if (skb_cow_head(skb, dev->needed_headroom))
681 			goto free_skb;
682 
683 		tnl_params = &tunnel->parms.iph;
684 	}
685 
686 	if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
687 		goto free_skb;
688 
689 	__gre_xmit(skb, dev, tnl_params, skb->protocol);
690 	return NETDEV_TX_OK;
691 
692 free_skb:
693 	kfree_skb(skb);
694 	dev->stats.tx_dropped++;
695 	return NETDEV_TX_OK;
696 }
697 
698 static netdev_tx_t erspan_xmit(struct sk_buff *skb,
699 			       struct net_device *dev)
700 {
701 	struct ip_tunnel *tunnel = netdev_priv(dev);
702 	bool truncate = false;
703 
704 	if (tunnel->collect_md) {
705 		erspan_fb_xmit(skb, dev, skb->protocol);
706 		return NETDEV_TX_OK;
707 	}
708 
709 	if (gre_handle_offloads(skb, false))
710 		goto free_skb;
711 
712 	if (skb_cow_head(skb, dev->needed_headroom))
713 		goto free_skb;
714 
715 	if (skb->len > dev->mtu + dev->hard_header_len) {
716 		pskb_trim(skb, dev->mtu + dev->hard_header_len);
717 		truncate = true;
718 	}
719 
720 	/* Push ERSPAN header */
721 	if (tunnel->erspan_ver == 1)
722 		erspan_build_header(skb, ntohl(tunnel->parms.o_key),
723 				    tunnel->index,
724 				    truncate, true);
725 	else
726 		erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key),
727 				       tunnel->dir, tunnel->hwid,
728 				       truncate, true);
729 
730 	tunnel->parms.o_flags &= ~TUNNEL_KEY;
731 	__gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_ERSPAN));
732 	return NETDEV_TX_OK;
733 
734 free_skb:
735 	kfree_skb(skb);
736 	dev->stats.tx_dropped++;
737 	return NETDEV_TX_OK;
738 }
739 
740 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
741 				struct net_device *dev)
742 {
743 	struct ip_tunnel *tunnel = netdev_priv(dev);
744 
745 	if (tunnel->collect_md) {
746 		gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
747 		return NETDEV_TX_OK;
748 	}
749 
750 	if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
751 		goto free_skb;
752 
753 	if (skb_cow_head(skb, dev->needed_headroom))
754 		goto free_skb;
755 
756 	__gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
757 	return NETDEV_TX_OK;
758 
759 free_skb:
760 	kfree_skb(skb);
761 	dev->stats.tx_dropped++;
762 	return NETDEV_TX_OK;
763 }
764 
765 static void ipgre_link_update(struct net_device *dev, bool set_mtu)
766 {
767 	struct ip_tunnel *tunnel = netdev_priv(dev);
768 	int len;
769 
770 	len = tunnel->tun_hlen;
771 	tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
772 	len = tunnel->tun_hlen - len;
773 	tunnel->hlen = tunnel->hlen + len;
774 
775 	dev->needed_headroom = dev->needed_headroom + len;
776 	if (set_mtu)
777 		dev->mtu = max_t(int, dev->mtu - len, 68);
778 
779 	if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
780 		if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
781 		    tunnel->encap.type == TUNNEL_ENCAP_NONE) {
782 			dev->features |= NETIF_F_GSO_SOFTWARE;
783 			dev->hw_features |= NETIF_F_GSO_SOFTWARE;
784 		} else {
785 			dev->features &= ~NETIF_F_GSO_SOFTWARE;
786 			dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
787 		}
788 		dev->features |= NETIF_F_LLTX;
789 	} else {
790 		dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
791 		dev->features &= ~(NETIF_F_LLTX | NETIF_F_GSO_SOFTWARE);
792 	}
793 }
794 
795 static int ipgre_tunnel_ioctl(struct net_device *dev,
796 			      struct ifreq *ifr, int cmd)
797 {
798 	struct ip_tunnel_parm p;
799 	int err;
800 
801 	if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
802 		return -EFAULT;
803 
804 	if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
805 		if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
806 		    p.iph.ihl != 5 || (p.iph.frag_off & htons(~IP_DF)) ||
807 		    ((p.i_flags | p.o_flags) & (GRE_VERSION | GRE_ROUTING)))
808 			return -EINVAL;
809 	}
810 
811 	p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
812 	p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
813 
814 	err = ip_tunnel_ioctl(dev, &p, cmd);
815 	if (err)
816 		return err;
817 
818 	if (cmd == SIOCCHGTUNNEL) {
819 		struct ip_tunnel *t = netdev_priv(dev);
820 
821 		t->parms.i_flags = p.i_flags;
822 		t->parms.o_flags = p.o_flags;
823 
824 		if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
825 			ipgre_link_update(dev, true);
826 	}
827 
828 	p.i_flags = gre_tnl_flags_to_gre_flags(p.i_flags);
829 	p.o_flags = gre_tnl_flags_to_gre_flags(p.o_flags);
830 
831 	if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
832 		return -EFAULT;
833 
834 	return 0;
835 }
836 
837 /* Nice toy. Unfortunately, useless in real life :-)
838    It allows to construct virtual multiprotocol broadcast "LAN"
839    over the Internet, provided multicast routing is tuned.
840 
841 
842    I have no idea was this bicycle invented before me,
843    so that I had to set ARPHRD_IPGRE to a random value.
844    I have an impression, that Cisco could make something similar,
845    but this feature is apparently missing in IOS<=11.2(8).
846 
847    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
848    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
849 
850    ping -t 255 224.66.66.66
851 
852    If nobody answers, mbone does not work.
853 
854    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
855    ip addr add 10.66.66.<somewhat>/24 dev Universe
856    ifconfig Universe up
857    ifconfig Universe add fe80::<Your_real_addr>/10
858    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
859    ftp 10.66.66.66
860    ...
861    ftp fec0:6666:6666::193.233.7.65
862    ...
863  */
864 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
865 			unsigned short type,
866 			const void *daddr, const void *saddr, unsigned int len)
867 {
868 	struct ip_tunnel *t = netdev_priv(dev);
869 	struct iphdr *iph;
870 	struct gre_base_hdr *greh;
871 
872 	iph = skb_push(skb, t->hlen + sizeof(*iph));
873 	greh = (struct gre_base_hdr *)(iph+1);
874 	greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
875 	greh->protocol = htons(type);
876 
877 	memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
878 
879 	/* Set the source hardware address. */
880 	if (saddr)
881 		memcpy(&iph->saddr, saddr, 4);
882 	if (daddr)
883 		memcpy(&iph->daddr, daddr, 4);
884 	if (iph->daddr)
885 		return t->hlen + sizeof(*iph);
886 
887 	return -(t->hlen + sizeof(*iph));
888 }
889 
890 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
891 {
892 	const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
893 	memcpy(haddr, &iph->saddr, 4);
894 	return 4;
895 }
896 
897 static const struct header_ops ipgre_header_ops = {
898 	.create	= ipgre_header,
899 	.parse	= ipgre_header_parse,
900 };
901 
902 #ifdef CONFIG_NET_IPGRE_BROADCAST
903 static int ipgre_open(struct net_device *dev)
904 {
905 	struct ip_tunnel *t = netdev_priv(dev);
906 
907 	if (ipv4_is_multicast(t->parms.iph.daddr)) {
908 		struct flowi4 fl4;
909 		struct rtable *rt;
910 
911 		rt = ip_route_output_gre(t->net, &fl4,
912 					 t->parms.iph.daddr,
913 					 t->parms.iph.saddr,
914 					 t->parms.o_key,
915 					 RT_TOS(t->parms.iph.tos),
916 					 t->parms.link);
917 		if (IS_ERR(rt))
918 			return -EADDRNOTAVAIL;
919 		dev = rt->dst.dev;
920 		ip_rt_put(rt);
921 		if (!__in_dev_get_rtnl(dev))
922 			return -EADDRNOTAVAIL;
923 		t->mlink = dev->ifindex;
924 		ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
925 	}
926 	return 0;
927 }
928 
929 static int ipgre_close(struct net_device *dev)
930 {
931 	struct ip_tunnel *t = netdev_priv(dev);
932 
933 	if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
934 		struct in_device *in_dev;
935 		in_dev = inetdev_by_index(t->net, t->mlink);
936 		if (in_dev)
937 			ip_mc_dec_group(in_dev, t->parms.iph.daddr);
938 	}
939 	return 0;
940 }
941 #endif
942 
943 static const struct net_device_ops ipgre_netdev_ops = {
944 	.ndo_init		= ipgre_tunnel_init,
945 	.ndo_uninit		= ip_tunnel_uninit,
946 #ifdef CONFIG_NET_IPGRE_BROADCAST
947 	.ndo_open		= ipgre_open,
948 	.ndo_stop		= ipgre_close,
949 #endif
950 	.ndo_start_xmit		= ipgre_xmit,
951 	.ndo_do_ioctl		= ipgre_tunnel_ioctl,
952 	.ndo_change_mtu		= ip_tunnel_change_mtu,
953 	.ndo_get_stats64	= ip_tunnel_get_stats64,
954 	.ndo_get_iflink		= ip_tunnel_get_iflink,
955 };
956 
957 #define GRE_FEATURES (NETIF_F_SG |		\
958 		      NETIF_F_FRAGLIST |	\
959 		      NETIF_F_HIGHDMA |		\
960 		      NETIF_F_HW_CSUM)
961 
962 static void ipgre_tunnel_setup(struct net_device *dev)
963 {
964 	dev->netdev_ops		= &ipgre_netdev_ops;
965 	dev->type		= ARPHRD_IPGRE;
966 	ip_tunnel_setup(dev, ipgre_net_id);
967 }
968 
969 static void __gre_tunnel_init(struct net_device *dev)
970 {
971 	struct ip_tunnel *tunnel;
972 	int t_hlen;
973 
974 	tunnel = netdev_priv(dev);
975 	tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
976 	tunnel->parms.iph.protocol = IPPROTO_GRE;
977 
978 	tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
979 
980 	t_hlen = tunnel->hlen + sizeof(struct iphdr);
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->features		|= GRE_FEATURES;
1300 	dev->hw_features	|= GRE_FEATURES;
1301 	dev->priv_flags		|= IFF_LIVE_ADDR_CHANGE;
1302 	netif_keep_dst(dev);
1303 
1304 	return ip_tunnel_init(dev);
1305 }
1306 
1307 static const struct net_device_ops erspan_netdev_ops = {
1308 	.ndo_init		= erspan_tunnel_init,
1309 	.ndo_uninit		= ip_tunnel_uninit,
1310 	.ndo_start_xmit		= erspan_xmit,
1311 	.ndo_set_mac_address	= eth_mac_addr,
1312 	.ndo_validate_addr	= eth_validate_addr,
1313 	.ndo_change_mtu		= ip_tunnel_change_mtu,
1314 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1315 	.ndo_get_iflink		= ip_tunnel_get_iflink,
1316 	.ndo_fill_metadata_dst	= gre_fill_metadata_dst,
1317 };
1318 
1319 static void ipgre_tap_setup(struct net_device *dev)
1320 {
1321 	ether_setup(dev);
1322 	dev->max_mtu = 0;
1323 	dev->netdev_ops	= &gre_tap_netdev_ops;
1324 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1325 	dev->priv_flags	|= IFF_LIVE_ADDR_CHANGE;
1326 	ip_tunnel_setup(dev, gre_tap_net_id);
1327 }
1328 
1329 bool is_gretap_dev(const struct net_device *dev)
1330 {
1331 	return dev->netdev_ops == &gre_tap_netdev_ops;
1332 }
1333 EXPORT_SYMBOL_GPL(is_gretap_dev);
1334 
1335 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1336 			 struct nlattr *tb[], struct nlattr *data[],
1337 			 struct netlink_ext_ack *extack)
1338 {
1339 	struct ip_tunnel_parm p;
1340 	struct ip_tunnel_encap ipencap;
1341 	__u32 fwmark = 0;
1342 	int err;
1343 
1344 	if (ipgre_netlink_encap_parms(data, &ipencap)) {
1345 		struct ip_tunnel *t = netdev_priv(dev);
1346 		err = ip_tunnel_encap_setup(t, &ipencap);
1347 
1348 		if (err < 0)
1349 			return err;
1350 	}
1351 
1352 	err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1353 	if (err < 0)
1354 		return err;
1355 	return ip_tunnel_newlink(dev, tb, &p, fwmark);
1356 }
1357 
1358 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1359 			    struct nlattr *data[],
1360 			    struct netlink_ext_ack *extack)
1361 {
1362 	struct ip_tunnel *t = netdev_priv(dev);
1363 	struct ip_tunnel_encap ipencap;
1364 	__u32 fwmark = t->fwmark;
1365 	struct ip_tunnel_parm p;
1366 	int err;
1367 
1368 	if (ipgre_netlink_encap_parms(data, &ipencap)) {
1369 		err = ip_tunnel_encap_setup(t, &ipencap);
1370 
1371 		if (err < 0)
1372 			return err;
1373 	}
1374 
1375 	err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1376 	if (err < 0)
1377 		return err;
1378 
1379 	err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1380 	if (err < 0)
1381 		return err;
1382 
1383 	t->parms.i_flags = p.i_flags;
1384 	t->parms.o_flags = p.o_flags;
1385 
1386 	if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
1387 		ipgre_link_update(dev, !tb[IFLA_MTU]);
1388 
1389 	return 0;
1390 }
1391 
1392 static size_t ipgre_get_size(const struct net_device *dev)
1393 {
1394 	return
1395 		/* IFLA_GRE_LINK */
1396 		nla_total_size(4) +
1397 		/* IFLA_GRE_IFLAGS */
1398 		nla_total_size(2) +
1399 		/* IFLA_GRE_OFLAGS */
1400 		nla_total_size(2) +
1401 		/* IFLA_GRE_IKEY */
1402 		nla_total_size(4) +
1403 		/* IFLA_GRE_OKEY */
1404 		nla_total_size(4) +
1405 		/* IFLA_GRE_LOCAL */
1406 		nla_total_size(4) +
1407 		/* IFLA_GRE_REMOTE */
1408 		nla_total_size(4) +
1409 		/* IFLA_GRE_TTL */
1410 		nla_total_size(1) +
1411 		/* IFLA_GRE_TOS */
1412 		nla_total_size(1) +
1413 		/* IFLA_GRE_PMTUDISC */
1414 		nla_total_size(1) +
1415 		/* IFLA_GRE_ENCAP_TYPE */
1416 		nla_total_size(2) +
1417 		/* IFLA_GRE_ENCAP_FLAGS */
1418 		nla_total_size(2) +
1419 		/* IFLA_GRE_ENCAP_SPORT */
1420 		nla_total_size(2) +
1421 		/* IFLA_GRE_ENCAP_DPORT */
1422 		nla_total_size(2) +
1423 		/* IFLA_GRE_COLLECT_METADATA */
1424 		nla_total_size(0) +
1425 		/* IFLA_GRE_IGNORE_DF */
1426 		nla_total_size(1) +
1427 		/* IFLA_GRE_FWMARK */
1428 		nla_total_size(4) +
1429 		/* IFLA_GRE_ERSPAN_INDEX */
1430 		nla_total_size(4) +
1431 		/* IFLA_GRE_ERSPAN_VER */
1432 		nla_total_size(1) +
1433 		/* IFLA_GRE_ERSPAN_DIR */
1434 		nla_total_size(1) +
1435 		/* IFLA_GRE_ERSPAN_HWID */
1436 		nla_total_size(2) +
1437 		0;
1438 }
1439 
1440 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1441 {
1442 	struct ip_tunnel *t = netdev_priv(dev);
1443 	struct ip_tunnel_parm *p = &t->parms;
1444 
1445 	if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1446 	    nla_put_be16(skb, IFLA_GRE_IFLAGS,
1447 			 gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1448 	    nla_put_be16(skb, IFLA_GRE_OFLAGS,
1449 			 gre_tnl_flags_to_gre_flags(p->o_flags)) ||
1450 	    nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1451 	    nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1452 	    nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1453 	    nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1454 	    nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1455 	    nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1456 	    nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1457 		       !!(p->iph.frag_off & htons(IP_DF))) ||
1458 	    nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1459 		goto nla_put_failure;
1460 
1461 	if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1462 			t->encap.type) ||
1463 	    nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1464 			 t->encap.sport) ||
1465 	    nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1466 			 t->encap.dport) ||
1467 	    nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1468 			t->encap.flags))
1469 		goto nla_put_failure;
1470 
1471 	if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1472 		goto nla_put_failure;
1473 
1474 	if (t->collect_md) {
1475 		if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1476 			goto nla_put_failure;
1477 	}
1478 
1479 	if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1480 		goto nla_put_failure;
1481 
1482 	if (t->erspan_ver == 1) {
1483 		if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1484 			goto nla_put_failure;
1485 	} else if (t->erspan_ver == 2) {
1486 		if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1487 			goto nla_put_failure;
1488 		if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1489 			goto nla_put_failure;
1490 	}
1491 
1492 	return 0;
1493 
1494 nla_put_failure:
1495 	return -EMSGSIZE;
1496 }
1497 
1498 static void erspan_setup(struct net_device *dev)
1499 {
1500 	ether_setup(dev);
1501 	dev->netdev_ops = &erspan_netdev_ops;
1502 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1503 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1504 	ip_tunnel_setup(dev, erspan_net_id);
1505 }
1506 
1507 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1508 	[IFLA_GRE_LINK]		= { .type = NLA_U32 },
1509 	[IFLA_GRE_IFLAGS]	= { .type = NLA_U16 },
1510 	[IFLA_GRE_OFLAGS]	= { .type = NLA_U16 },
1511 	[IFLA_GRE_IKEY]		= { .type = NLA_U32 },
1512 	[IFLA_GRE_OKEY]		= { .type = NLA_U32 },
1513 	[IFLA_GRE_LOCAL]	= { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1514 	[IFLA_GRE_REMOTE]	= { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1515 	[IFLA_GRE_TTL]		= { .type = NLA_U8 },
1516 	[IFLA_GRE_TOS]		= { .type = NLA_U8 },
1517 	[IFLA_GRE_PMTUDISC]	= { .type = NLA_U8 },
1518 	[IFLA_GRE_ENCAP_TYPE]	= { .type = NLA_U16 },
1519 	[IFLA_GRE_ENCAP_FLAGS]	= { .type = NLA_U16 },
1520 	[IFLA_GRE_ENCAP_SPORT]	= { .type = NLA_U16 },
1521 	[IFLA_GRE_ENCAP_DPORT]	= { .type = NLA_U16 },
1522 	[IFLA_GRE_COLLECT_METADATA]	= { .type = NLA_FLAG },
1523 	[IFLA_GRE_IGNORE_DF]	= { .type = NLA_U8 },
1524 	[IFLA_GRE_FWMARK]	= { .type = NLA_U32 },
1525 	[IFLA_GRE_ERSPAN_INDEX]	= { .type = NLA_U32 },
1526 	[IFLA_GRE_ERSPAN_VER]	= { .type = NLA_U8 },
1527 	[IFLA_GRE_ERSPAN_DIR]	= { .type = NLA_U8 },
1528 	[IFLA_GRE_ERSPAN_HWID]	= { .type = NLA_U16 },
1529 };
1530 
1531 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1532 	.kind		= "gre",
1533 	.maxtype	= IFLA_GRE_MAX,
1534 	.policy		= ipgre_policy,
1535 	.priv_size	= sizeof(struct ip_tunnel),
1536 	.setup		= ipgre_tunnel_setup,
1537 	.validate	= ipgre_tunnel_validate,
1538 	.newlink	= ipgre_newlink,
1539 	.changelink	= ipgre_changelink,
1540 	.dellink	= ip_tunnel_dellink,
1541 	.get_size	= ipgre_get_size,
1542 	.fill_info	= ipgre_fill_info,
1543 	.get_link_net	= ip_tunnel_get_link_net,
1544 };
1545 
1546 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1547 	.kind		= "gretap",
1548 	.maxtype	= IFLA_GRE_MAX,
1549 	.policy		= ipgre_policy,
1550 	.priv_size	= sizeof(struct ip_tunnel),
1551 	.setup		= ipgre_tap_setup,
1552 	.validate	= ipgre_tap_validate,
1553 	.newlink	= ipgre_newlink,
1554 	.changelink	= ipgre_changelink,
1555 	.dellink	= ip_tunnel_dellink,
1556 	.get_size	= ipgre_get_size,
1557 	.fill_info	= ipgre_fill_info,
1558 	.get_link_net	= ip_tunnel_get_link_net,
1559 };
1560 
1561 static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1562 	.kind		= "erspan",
1563 	.maxtype	= IFLA_GRE_MAX,
1564 	.policy		= ipgre_policy,
1565 	.priv_size	= sizeof(struct ip_tunnel),
1566 	.setup		= erspan_setup,
1567 	.validate	= erspan_validate,
1568 	.newlink	= ipgre_newlink,
1569 	.changelink	= ipgre_changelink,
1570 	.dellink	= ip_tunnel_dellink,
1571 	.get_size	= ipgre_get_size,
1572 	.fill_info	= ipgre_fill_info,
1573 	.get_link_net	= ip_tunnel_get_link_net,
1574 };
1575 
1576 struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1577 					u8 name_assign_type)
1578 {
1579 	struct nlattr *tb[IFLA_MAX + 1];
1580 	struct net_device *dev;
1581 	LIST_HEAD(list_kill);
1582 	struct ip_tunnel *t;
1583 	int err;
1584 
1585 	memset(&tb, 0, sizeof(tb));
1586 
1587 	dev = rtnl_create_link(net, name, name_assign_type,
1588 			       &ipgre_tap_ops, tb);
1589 	if (IS_ERR(dev))
1590 		return dev;
1591 
1592 	/* Configure flow based GRE device. */
1593 	t = netdev_priv(dev);
1594 	t->collect_md = true;
1595 
1596 	err = ipgre_newlink(net, dev, tb, NULL, NULL);
1597 	if (err < 0) {
1598 		free_netdev(dev);
1599 		return ERR_PTR(err);
1600 	}
1601 
1602 	/* openvswitch users expect packet sizes to be unrestricted,
1603 	 * so set the largest MTU we can.
1604 	 */
1605 	err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1606 	if (err)
1607 		goto out;
1608 
1609 	err = rtnl_configure_link(dev, NULL);
1610 	if (err < 0)
1611 		goto out;
1612 
1613 	return dev;
1614 out:
1615 	ip_tunnel_dellink(dev, &list_kill);
1616 	unregister_netdevice_many(&list_kill);
1617 	return ERR_PTR(err);
1618 }
1619 EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1620 
1621 static int __net_init ipgre_tap_init_net(struct net *net)
1622 {
1623 	return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1624 }
1625 
1626 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net)
1627 {
1628 	ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops);
1629 }
1630 
1631 static struct pernet_operations ipgre_tap_net_ops = {
1632 	.init = ipgre_tap_init_net,
1633 	.exit_batch = ipgre_tap_exit_batch_net,
1634 	.id   = &gre_tap_net_id,
1635 	.size = sizeof(struct ip_tunnel_net),
1636 };
1637 
1638 static int __net_init erspan_init_net(struct net *net)
1639 {
1640 	return ip_tunnel_init_net(net, erspan_net_id,
1641 				  &erspan_link_ops, "erspan0");
1642 }
1643 
1644 static void __net_exit erspan_exit_batch_net(struct list_head *net_list)
1645 {
1646 	ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops);
1647 }
1648 
1649 static struct pernet_operations erspan_net_ops = {
1650 	.init = erspan_init_net,
1651 	.exit_batch = erspan_exit_batch_net,
1652 	.id   = &erspan_net_id,
1653 	.size = sizeof(struct ip_tunnel_net),
1654 };
1655 
1656 static int __init ipgre_init(void)
1657 {
1658 	int err;
1659 
1660 	pr_info("GRE over IPv4 tunneling driver\n");
1661 
1662 	err = register_pernet_device(&ipgre_net_ops);
1663 	if (err < 0)
1664 		return err;
1665 
1666 	err = register_pernet_device(&ipgre_tap_net_ops);
1667 	if (err < 0)
1668 		goto pnet_tap_failed;
1669 
1670 	err = register_pernet_device(&erspan_net_ops);
1671 	if (err < 0)
1672 		goto pnet_erspan_failed;
1673 
1674 	err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1675 	if (err < 0) {
1676 		pr_info("%s: can't add protocol\n", __func__);
1677 		goto add_proto_failed;
1678 	}
1679 
1680 	err = rtnl_link_register(&ipgre_link_ops);
1681 	if (err < 0)
1682 		goto rtnl_link_failed;
1683 
1684 	err = rtnl_link_register(&ipgre_tap_ops);
1685 	if (err < 0)
1686 		goto tap_ops_failed;
1687 
1688 	err = rtnl_link_register(&erspan_link_ops);
1689 	if (err < 0)
1690 		goto erspan_link_failed;
1691 
1692 	return 0;
1693 
1694 erspan_link_failed:
1695 	rtnl_link_unregister(&ipgre_tap_ops);
1696 tap_ops_failed:
1697 	rtnl_link_unregister(&ipgre_link_ops);
1698 rtnl_link_failed:
1699 	gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1700 add_proto_failed:
1701 	unregister_pernet_device(&erspan_net_ops);
1702 pnet_erspan_failed:
1703 	unregister_pernet_device(&ipgre_tap_net_ops);
1704 pnet_tap_failed:
1705 	unregister_pernet_device(&ipgre_net_ops);
1706 	return err;
1707 }
1708 
1709 static void __exit ipgre_fini(void)
1710 {
1711 	rtnl_link_unregister(&ipgre_tap_ops);
1712 	rtnl_link_unregister(&ipgre_link_ops);
1713 	rtnl_link_unregister(&erspan_link_ops);
1714 	gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1715 	unregister_pernet_device(&ipgre_tap_net_ops);
1716 	unregister_pernet_device(&ipgre_net_ops);
1717 	unregister_pernet_device(&erspan_net_ops);
1718 }
1719 
1720 module_init(ipgre_init);
1721 module_exit(ipgre_fini);
1722 MODULE_LICENSE("GPL");
1723 MODULE_ALIAS_RTNL_LINK("gre");
1724 MODULE_ALIAS_RTNL_LINK("gretap");
1725 MODULE_ALIAS_RTNL_LINK("erspan");
1726 MODULE_ALIAS_NETDEV("gre0");
1727 MODULE_ALIAS_NETDEV("gretap0");
1728 MODULE_ALIAS_NETDEV("erspan0");
1729