xref: /openbmc/linux/net/ipv4/ip_gre.c (revision dc6a81c3)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	Linux NET3:	GRE over IP protocol decoder.
4  *
5  *	Authors: Alexey Kuznetsov (kuznet@ms2.inr.ac.ru)
6  */
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
10 #include <linux/capability.h>
11 #include <linux/module.h>
12 #include <linux/types.h>
13 #include <linux/kernel.h>
14 #include <linux/slab.h>
15 #include <linux/uaccess.h>
16 #include <linux/skbuff.h>
17 #include <linux/netdevice.h>
18 #include <linux/in.h>
19 #include <linux/tcp.h>
20 #include <linux/udp.h>
21 #include <linux/if_arp.h>
22 #include <linux/if_vlan.h>
23 #include <linux/init.h>
24 #include <linux/in6.h>
25 #include <linux/inetdevice.h>
26 #include <linux/igmp.h>
27 #include <linux/netfilter_ipv4.h>
28 #include <linux/etherdevice.h>
29 #include <linux/if_ether.h>
30 
31 #include <net/sock.h>
32 #include <net/ip.h>
33 #include <net/icmp.h>
34 #include <net/protocol.h>
35 #include <net/ip_tunnels.h>
36 #include <net/arp.h>
37 #include <net/checksum.h>
38 #include <net/dsfield.h>
39 #include <net/inet_ecn.h>
40 #include <net/xfrm.h>
41 #include <net/net_namespace.h>
42 #include <net/netns/generic.h>
43 #include <net/rtnetlink.h>
44 #include <net/gre.h>
45 #include <net/dst_metadata.h>
46 #include <net/erspan.h>
47 
48 /*
49    Problems & solutions
50    --------------------
51 
52    1. The most important issue is detecting local dead loops.
53    They would cause complete host lockup in transmit, which
54    would be "resolved" by stack overflow or, if queueing is enabled,
55    with infinite looping in net_bh.
56 
57    We cannot track such dead loops during route installation,
58    it is infeasible task. The most general solutions would be
59    to keep skb->encapsulation counter (sort of local ttl),
60    and silently drop packet when it expires. It is a good
61    solution, but it supposes maintaining new variable in ALL
62    skb, even if no tunneling is used.
63 
64    Current solution: xmit_recursion breaks dead loops. This is a percpu
65    counter, since when we enter the first ndo_xmit(), cpu migration is
66    forbidden. We force an exit if this counter reaches RECURSION_LIMIT
67 
68    2. Networking dead loops would not kill routers, but would really
69    kill network. IP hop limit plays role of "t->recursion" in this case,
70    if we copy it from packet being encapsulated to upper header.
71    It is very good solution, but it introduces two problems:
72 
73    - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
74      do not work over tunnels.
75    - traceroute does not work. I planned to relay ICMP from tunnel,
76      so that this problem would be solved and traceroute output
77      would even more informative. This idea appeared to be wrong:
78      only Linux complies to rfc1812 now (yes, guys, Linux is the only
79      true router now :-)), all routers (at least, in neighbourhood of mine)
80      return only 8 bytes of payload. It is the end.
81 
82    Hence, if we want that OSPF worked or traceroute said something reasonable,
83    we should search for another solution.
84 
85    One of them is to parse packet trying to detect inner encapsulation
86    made by our node. It is difficult or even impossible, especially,
87    taking into account fragmentation. TO be short, ttl is not solution at all.
88 
89    Current solution: The solution was UNEXPECTEDLY SIMPLE.
90    We force DF flag on tunnels with preconfigured hop limit,
91    that is ALL. :-) Well, it does not remove the problem completely,
92    but exponential growth of network traffic is changed to linear
93    (branches, that exceed pmtu are pruned) and tunnel mtu
94    rapidly degrades to value <68, where looping stops.
95    Yes, it is not good if there exists a router in the loop,
96    which does not force DF, even when encapsulating packets have DF set.
97    But it is not our problem! Nobody could accuse us, we made
98    all that we could make. Even if it is your gated who injected
99    fatal route to network, even if it were you who configured
100    fatal static route: you are innocent. :-)
101 
102    Alexey Kuznetsov.
103  */
104 
105 static bool log_ecn_error = true;
106 module_param(log_ecn_error, bool, 0644);
107 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
108 
109 static struct rtnl_link_ops ipgre_link_ops __read_mostly;
110 static int ipgre_tunnel_init(struct net_device *dev);
111 static void erspan_build_header(struct sk_buff *skb,
112 				u32 id, u32 index,
113 				bool truncate, bool is_ipv4);
114 
115 static unsigned int ipgre_net_id __read_mostly;
116 static unsigned int gre_tap_net_id __read_mostly;
117 static unsigned int erspan_net_id __read_mostly;
118 
119 static int ipgre_err(struct sk_buff *skb, u32 info,
120 		     const struct tnl_ptk_info *tpi)
121 {
122 
123 	/* All the routers (except for Linux) return only
124 	   8 bytes of packet payload. It means, that precise relaying of
125 	   ICMP in the real Internet is absolutely infeasible.
126 
127 	   Moreover, Cisco "wise men" put GRE key to the third word
128 	   in GRE header. It makes impossible maintaining even soft
129 	   state for keyed GRE tunnels with enabled checksum. Tell
130 	   them "thank you".
131 
132 	   Well, I wonder, rfc1812 was written by Cisco employee,
133 	   what the hell these idiots break standards established
134 	   by themselves???
135 	   */
136 	struct net *net = dev_net(skb->dev);
137 	struct ip_tunnel_net *itn;
138 	const struct iphdr *iph;
139 	const int type = icmp_hdr(skb)->type;
140 	const int code = icmp_hdr(skb)->code;
141 	unsigned int data_len = 0;
142 	struct ip_tunnel *t;
143 
144 	if (tpi->proto == htons(ETH_P_TEB))
145 		itn = net_generic(net, gre_tap_net_id);
146 	else if (tpi->proto == htons(ETH_P_ERSPAN) ||
147 		 tpi->proto == htons(ETH_P_ERSPAN2))
148 		itn = net_generic(net, erspan_net_id);
149 	else
150 		itn = net_generic(net, ipgre_net_id);
151 
152 	iph = (const struct iphdr *)(icmp_hdr(skb) + 1);
153 	t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
154 			     iph->daddr, iph->saddr, tpi->key);
155 
156 	if (!t)
157 		return -ENOENT;
158 
159 	switch (type) {
160 	default:
161 	case ICMP_PARAMETERPROB:
162 		return 0;
163 
164 	case ICMP_DEST_UNREACH:
165 		switch (code) {
166 		case ICMP_SR_FAILED:
167 		case ICMP_PORT_UNREACH:
168 			/* Impossible event. */
169 			return 0;
170 		default:
171 			/* All others are translated to HOST_UNREACH.
172 			   rfc2003 contains "deep thoughts" about NET_UNREACH,
173 			   I believe they are just ether pollution. --ANK
174 			 */
175 			break;
176 		}
177 		break;
178 
179 	case ICMP_TIME_EXCEEDED:
180 		if (code != ICMP_EXC_TTL)
181 			return 0;
182 		data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */
183 		break;
184 
185 	case ICMP_REDIRECT:
186 		break;
187 	}
188 
189 #if IS_ENABLED(CONFIG_IPV6)
190        if (tpi->proto == htons(ETH_P_IPV6) &&
191            !ip6_err_gen_icmpv6_unreach(skb, iph->ihl * 4 + tpi->hdr_len,
192 				       type, data_len))
193                return 0;
194 #endif
195 
196 	if (t->parms.iph.daddr == 0 ||
197 	    ipv4_is_multicast(t->parms.iph.daddr))
198 		return 0;
199 
200 	if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
201 		return 0;
202 
203 	if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
204 		t->err_count++;
205 	else
206 		t->err_count = 1;
207 	t->err_time = jiffies;
208 
209 	return 0;
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 
233 	if (gre_parse_header(skb, &tpi, NULL, htons(ETH_P_IP),
234 			     iph->ihl * 4) < 0)
235 		return;
236 
237 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
238 		ipv4_update_pmtu(skb, dev_net(skb->dev), info,
239 				 skb->dev->ifindex, IPPROTO_GRE);
240 		return;
241 	}
242 	if (type == ICMP_REDIRECT) {
243 		ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex,
244 			      IPPROTO_GRE);
245 		return;
246 	}
247 
248 	ipgre_err(skb, info, &tpi);
249 }
250 
251 static int erspan_rcv(struct sk_buff *skb, struct tnl_ptk_info *tpi,
252 		      int gre_hdr_len)
253 {
254 	struct net *net = dev_net(skb->dev);
255 	struct metadata_dst *tun_dst = NULL;
256 	struct erspan_base_hdr *ershdr;
257 	struct ip_tunnel_net *itn;
258 	struct ip_tunnel *tunnel;
259 	const struct iphdr *iph;
260 	struct erspan_md2 *md2;
261 	int ver;
262 	int len;
263 
264 	itn = net_generic(net, erspan_net_id);
265 
266 	iph = ip_hdr(skb);
267 	ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
268 	ver = ershdr->ver;
269 
270 	tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex,
271 				  tpi->flags | TUNNEL_KEY,
272 				  iph->saddr, iph->daddr, tpi->key);
273 
274 	if (tunnel) {
275 		len = gre_hdr_len + erspan_hdr_len(ver);
276 		if (unlikely(!pskb_may_pull(skb, len)))
277 			return PACKET_REJECT;
278 
279 		if (__iptunnel_pull_header(skb,
280 					   len,
281 					   htons(ETH_P_TEB),
282 					   false, false) < 0)
283 			goto drop;
284 
285 		if (tunnel->collect_md) {
286 			struct erspan_metadata *pkt_md, *md;
287 			struct ip_tunnel_info *info;
288 			unsigned char *gh;
289 			__be64 tun_id;
290 			__be16 flags;
291 
292 			tpi->flags |= TUNNEL_KEY;
293 			flags = tpi->flags;
294 			tun_id = key32_to_tunnel_id(tpi->key);
295 
296 			tun_dst = ip_tun_rx_dst(skb, flags,
297 						tun_id, sizeof(*md));
298 			if (!tun_dst)
299 				return PACKET_REJECT;
300 
301 			/* skb can be uncloned in __iptunnel_pull_header, so
302 			 * old pkt_md is no longer valid and we need to reset
303 			 * it
304 			 */
305 			gh = skb_network_header(skb) +
306 			     skb_network_header_len(skb);
307 			pkt_md = (struct erspan_metadata *)(gh + gre_hdr_len +
308 							    sizeof(*ershdr));
309 			md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
310 			md->version = ver;
311 			md2 = &md->u.md2;
312 			memcpy(md2, pkt_md, ver == 1 ? ERSPAN_V1_MDSIZE :
313 						       ERSPAN_V2_MDSIZE);
314 
315 			info = &tun_dst->u.tun_info;
316 			info->key.tun_flags |= TUNNEL_ERSPAN_OPT;
317 			info->options_len = sizeof(*md);
318 		}
319 
320 		skb_reset_mac_header(skb);
321 		ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
322 		return PACKET_RCVD;
323 	}
324 	return PACKET_REJECT;
325 
326 drop:
327 	kfree_skb(skb);
328 	return PACKET_RCVD;
329 }
330 
331 static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
332 		       struct ip_tunnel_net *itn, int hdr_len, bool raw_proto)
333 {
334 	struct metadata_dst *tun_dst = NULL;
335 	const struct iphdr *iph;
336 	struct ip_tunnel *tunnel;
337 
338 	iph = ip_hdr(skb);
339 	tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
340 				  iph->saddr, iph->daddr, tpi->key);
341 
342 	if (tunnel) {
343 		const struct iphdr *tnl_params;
344 
345 		if (__iptunnel_pull_header(skb, hdr_len, tpi->proto,
346 					   raw_proto, false) < 0)
347 			goto drop;
348 
349 		if (tunnel->dev->type != ARPHRD_NONE)
350 			skb_pop_mac_header(skb);
351 		else
352 			skb_reset_mac_header(skb);
353 
354 		tnl_params = &tunnel->parms.iph;
355 		if (tunnel->collect_md || tnl_params->daddr == 0) {
356 			__be16 flags;
357 			__be64 tun_id;
358 
359 			flags = tpi->flags & (TUNNEL_CSUM | TUNNEL_KEY);
360 			tun_id = key32_to_tunnel_id(tpi->key);
361 			tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
362 			if (!tun_dst)
363 				return PACKET_REJECT;
364 		}
365 
366 		ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
367 		return PACKET_RCVD;
368 	}
369 	return PACKET_NEXT;
370 
371 drop:
372 	kfree_skb(skb);
373 	return PACKET_RCVD;
374 }
375 
376 static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
377 		     int hdr_len)
378 {
379 	struct net *net = dev_net(skb->dev);
380 	struct ip_tunnel_net *itn;
381 	int res;
382 
383 	if (tpi->proto == htons(ETH_P_TEB))
384 		itn = net_generic(net, gre_tap_net_id);
385 	else
386 		itn = net_generic(net, ipgre_net_id);
387 
388 	res = __ipgre_rcv(skb, tpi, itn, hdr_len, false);
389 	if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) {
390 		/* ipgre tunnels in collect metadata mode should receive
391 		 * also ETH_P_TEB traffic.
392 		 */
393 		itn = net_generic(net, ipgre_net_id);
394 		res = __ipgre_rcv(skb, tpi, itn, hdr_len, true);
395 	}
396 	return res;
397 }
398 
399 static int gre_rcv(struct sk_buff *skb)
400 {
401 	struct tnl_ptk_info tpi;
402 	bool csum_err = false;
403 	int hdr_len;
404 
405 #ifdef CONFIG_NET_IPGRE_BROADCAST
406 	if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
407 		/* Looped back packet, drop it! */
408 		if (rt_is_output_route(skb_rtable(skb)))
409 			goto drop;
410 	}
411 #endif
412 
413 	hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0);
414 	if (hdr_len < 0)
415 		goto drop;
416 
417 	if (unlikely(tpi.proto == htons(ETH_P_ERSPAN) ||
418 		     tpi.proto == htons(ETH_P_ERSPAN2))) {
419 		if (erspan_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
420 			return 0;
421 		goto out;
422 	}
423 
424 	if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
425 		return 0;
426 
427 out:
428 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
429 drop:
430 	kfree_skb(skb);
431 	return 0;
432 }
433 
434 static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
435 		       const struct iphdr *tnl_params,
436 		       __be16 proto)
437 {
438 	struct ip_tunnel *tunnel = netdev_priv(dev);
439 
440 	if (tunnel->parms.o_flags & TUNNEL_SEQ)
441 		tunnel->o_seqno++;
442 
443 	/* Push GRE header. */
444 	gre_build_header(skb, tunnel->tun_hlen,
445 			 tunnel->parms.o_flags, proto, tunnel->parms.o_key,
446 			 htonl(tunnel->o_seqno));
447 
448 	ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
449 }
450 
451 static int gre_handle_offloads(struct sk_buff *skb, bool csum)
452 {
453 	return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
454 }
455 
456 static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev,
457 			__be16 proto)
458 {
459 	struct ip_tunnel *tunnel = netdev_priv(dev);
460 	struct ip_tunnel_info *tun_info;
461 	const struct ip_tunnel_key *key;
462 	int tunnel_hlen;
463 	__be16 flags;
464 
465 	tun_info = skb_tunnel_info(skb);
466 	if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
467 		     ip_tunnel_info_af(tun_info) != AF_INET))
468 		goto err_free_skb;
469 
470 	key = &tun_info->key;
471 	tunnel_hlen = gre_calc_hlen(key->tun_flags);
472 
473 	if (skb_cow_head(skb, dev->needed_headroom))
474 		goto err_free_skb;
475 
476 	/* Push Tunnel header. */
477 	if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM)))
478 		goto err_free_skb;
479 
480 	flags = tun_info->key.tun_flags &
481 		(TUNNEL_CSUM | TUNNEL_KEY | TUNNEL_SEQ);
482 	gre_build_header(skb, tunnel_hlen, flags, proto,
483 			 tunnel_id_to_key32(tun_info->key.tun_id),
484 			 (flags & TUNNEL_SEQ) ? htonl(tunnel->o_seqno++) : 0);
485 
486 	ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen);
487 
488 	return;
489 
490 err_free_skb:
491 	kfree_skb(skb);
492 	dev->stats.tx_dropped++;
493 }
494 
495 static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev)
496 {
497 	struct ip_tunnel *tunnel = netdev_priv(dev);
498 	struct ip_tunnel_info *tun_info;
499 	const struct ip_tunnel_key *key;
500 	struct erspan_metadata *md;
501 	bool truncate = false;
502 	__be16 proto;
503 	int tunnel_hlen;
504 	int version;
505 	int nhoff;
506 	int thoff;
507 
508 	tun_info = skb_tunnel_info(skb);
509 	if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
510 		     ip_tunnel_info_af(tun_info) != AF_INET))
511 		goto err_free_skb;
512 
513 	key = &tun_info->key;
514 	if (!(tun_info->key.tun_flags & TUNNEL_ERSPAN_OPT))
515 		goto err_free_skb;
516 	if (tun_info->options_len < sizeof(*md))
517 		goto err_free_skb;
518 	md = ip_tunnel_info_opts(tun_info);
519 
520 	/* ERSPAN has fixed 8 byte GRE header */
521 	version = md->version;
522 	tunnel_hlen = 8 + erspan_hdr_len(version);
523 
524 	if (skb_cow_head(skb, dev->needed_headroom))
525 		goto err_free_skb;
526 
527 	if (gre_handle_offloads(skb, false))
528 		goto err_free_skb;
529 
530 	if (skb->len > dev->mtu + dev->hard_header_len) {
531 		pskb_trim(skb, dev->mtu + dev->hard_header_len);
532 		truncate = true;
533 	}
534 
535 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
536 	if (skb->protocol == htons(ETH_P_IP) &&
537 	    (ntohs(ip_hdr(skb)->tot_len) > skb->len - nhoff))
538 		truncate = true;
539 
540 	thoff = skb_transport_header(skb) - skb_mac_header(skb);
541 	if (skb->protocol == htons(ETH_P_IPV6) &&
542 	    (ntohs(ipv6_hdr(skb)->payload_len) > skb->len - thoff))
543 		truncate = true;
544 
545 	if (version == 1) {
546 		erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)),
547 				    ntohl(md->u.index), truncate, true);
548 		proto = htons(ETH_P_ERSPAN);
549 	} else if (version == 2) {
550 		erspan_build_header_v2(skb,
551 				       ntohl(tunnel_id_to_key32(key->tun_id)),
552 				       md->u.md2.dir,
553 				       get_hwid(&md->u.md2),
554 				       truncate, true);
555 		proto = htons(ETH_P_ERSPAN2);
556 	} else {
557 		goto err_free_skb;
558 	}
559 
560 	gre_build_header(skb, 8, TUNNEL_SEQ,
561 			 proto, 0, htonl(tunnel->o_seqno++));
562 
563 	ip_md_tunnel_xmit(skb, dev, IPPROTO_GRE, tunnel_hlen);
564 
565 	return;
566 
567 err_free_skb:
568 	kfree_skb(skb);
569 	dev->stats.tx_dropped++;
570 }
571 
572 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
573 {
574 	struct ip_tunnel_info *info = skb_tunnel_info(skb);
575 	const struct ip_tunnel_key *key;
576 	struct rtable *rt;
577 	struct flowi4 fl4;
578 
579 	if (ip_tunnel_info_af(info) != AF_INET)
580 		return -EINVAL;
581 
582 	key = &info->key;
583 	ip_tunnel_init_flow(&fl4, IPPROTO_GRE, key->u.ipv4.dst, key->u.ipv4.src,
584 			    tunnel_id_to_key32(key->tun_id), key->tos, 0,
585 			    skb->mark, skb_get_hash(skb));
586 	rt = ip_route_output_key(dev_net(dev), &fl4);
587 	if (IS_ERR(rt))
588 		return PTR_ERR(rt);
589 
590 	ip_rt_put(rt);
591 	info->key.u.ipv4.src = fl4.saddr;
592 	return 0;
593 }
594 
595 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
596 			      struct net_device *dev)
597 {
598 	struct ip_tunnel *tunnel = netdev_priv(dev);
599 	const struct iphdr *tnl_params;
600 
601 	if (!pskb_inet_may_pull(skb))
602 		goto free_skb;
603 
604 	if (tunnel->collect_md) {
605 		gre_fb_xmit(skb, dev, skb->protocol);
606 		return NETDEV_TX_OK;
607 	}
608 
609 	if (dev->header_ops) {
610 		/* Need space for new headers */
611 		if (skb_cow_head(skb, dev->needed_headroom -
612 				      (tunnel->hlen + sizeof(struct iphdr))))
613 			goto free_skb;
614 
615 		tnl_params = (const struct iphdr *)skb->data;
616 
617 		/* Pull skb since ip_tunnel_xmit() needs skb->data pointing
618 		 * to gre header.
619 		 */
620 		skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
621 		skb_reset_mac_header(skb);
622 	} else {
623 		if (skb_cow_head(skb, dev->needed_headroom))
624 			goto free_skb;
625 
626 		tnl_params = &tunnel->parms.iph;
627 	}
628 
629 	if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
630 		goto free_skb;
631 
632 	__gre_xmit(skb, dev, tnl_params, skb->protocol);
633 	return NETDEV_TX_OK;
634 
635 free_skb:
636 	kfree_skb(skb);
637 	dev->stats.tx_dropped++;
638 	return NETDEV_TX_OK;
639 }
640 
641 static netdev_tx_t erspan_xmit(struct sk_buff *skb,
642 			       struct net_device *dev)
643 {
644 	struct ip_tunnel *tunnel = netdev_priv(dev);
645 	bool truncate = false;
646 	__be16 proto;
647 
648 	if (!pskb_inet_may_pull(skb))
649 		goto free_skb;
650 
651 	if (tunnel->collect_md) {
652 		erspan_fb_xmit(skb, dev);
653 		return NETDEV_TX_OK;
654 	}
655 
656 	if (gre_handle_offloads(skb, false))
657 		goto free_skb;
658 
659 	if (skb_cow_head(skb, dev->needed_headroom))
660 		goto free_skb;
661 
662 	if (skb->len > dev->mtu + dev->hard_header_len) {
663 		pskb_trim(skb, dev->mtu + dev->hard_header_len);
664 		truncate = true;
665 	}
666 
667 	/* Push ERSPAN header */
668 	if (tunnel->erspan_ver == 1) {
669 		erspan_build_header(skb, ntohl(tunnel->parms.o_key),
670 				    tunnel->index,
671 				    truncate, true);
672 		proto = htons(ETH_P_ERSPAN);
673 	} else if (tunnel->erspan_ver == 2) {
674 		erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key),
675 				       tunnel->dir, tunnel->hwid,
676 				       truncate, true);
677 		proto = htons(ETH_P_ERSPAN2);
678 	} else {
679 		goto free_skb;
680 	}
681 
682 	tunnel->parms.o_flags &= ~TUNNEL_KEY;
683 	__gre_xmit(skb, dev, &tunnel->parms.iph, proto);
684 	return NETDEV_TX_OK;
685 
686 free_skb:
687 	kfree_skb(skb);
688 	dev->stats.tx_dropped++;
689 	return NETDEV_TX_OK;
690 }
691 
692 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
693 				struct net_device *dev)
694 {
695 	struct ip_tunnel *tunnel = netdev_priv(dev);
696 
697 	if (!pskb_inet_may_pull(skb))
698 		goto free_skb;
699 
700 	if (tunnel->collect_md) {
701 		gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
702 		return NETDEV_TX_OK;
703 	}
704 
705 	if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
706 		goto free_skb;
707 
708 	if (skb_cow_head(skb, dev->needed_headroom))
709 		goto free_skb;
710 
711 	__gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
712 	return NETDEV_TX_OK;
713 
714 free_skb:
715 	kfree_skb(skb);
716 	dev->stats.tx_dropped++;
717 	return NETDEV_TX_OK;
718 }
719 
720 static void ipgre_link_update(struct net_device *dev, bool set_mtu)
721 {
722 	struct ip_tunnel *tunnel = netdev_priv(dev);
723 	int len;
724 
725 	len = tunnel->tun_hlen;
726 	tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
727 	len = tunnel->tun_hlen - len;
728 	tunnel->hlen = tunnel->hlen + len;
729 
730 	dev->needed_headroom = dev->needed_headroom + len;
731 	if (set_mtu)
732 		dev->mtu = max_t(int, dev->mtu - len, 68);
733 
734 	if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
735 		if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
736 		    tunnel->encap.type == TUNNEL_ENCAP_NONE) {
737 			dev->features |= NETIF_F_GSO_SOFTWARE;
738 			dev->hw_features |= NETIF_F_GSO_SOFTWARE;
739 		} else {
740 			dev->features &= ~NETIF_F_GSO_SOFTWARE;
741 			dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
742 		}
743 		dev->features |= NETIF_F_LLTX;
744 	} else {
745 		dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
746 		dev->features &= ~(NETIF_F_LLTX | NETIF_F_GSO_SOFTWARE);
747 	}
748 }
749 
750 static int ipgre_tunnel_ioctl(struct net_device *dev,
751 			      struct ifreq *ifr, int cmd)
752 {
753 	struct ip_tunnel_parm p;
754 	int err;
755 
756 	if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
757 		return -EFAULT;
758 
759 	if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
760 		if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
761 		    p.iph.ihl != 5 || (p.iph.frag_off & htons(~IP_DF)) ||
762 		    ((p.i_flags | p.o_flags) & (GRE_VERSION | GRE_ROUTING)))
763 			return -EINVAL;
764 	}
765 
766 	p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
767 	p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
768 
769 	err = ip_tunnel_ioctl(dev, &p, cmd);
770 	if (err)
771 		return err;
772 
773 	if (cmd == SIOCCHGTUNNEL) {
774 		struct ip_tunnel *t = netdev_priv(dev);
775 
776 		t->parms.i_flags = p.i_flags;
777 		t->parms.o_flags = p.o_flags;
778 
779 		if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
780 			ipgre_link_update(dev, true);
781 	}
782 
783 	p.i_flags = gre_tnl_flags_to_gre_flags(p.i_flags);
784 	p.o_flags = gre_tnl_flags_to_gre_flags(p.o_flags);
785 
786 	if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
787 		return -EFAULT;
788 
789 	return 0;
790 }
791 
792 /* Nice toy. Unfortunately, useless in real life :-)
793    It allows to construct virtual multiprotocol broadcast "LAN"
794    over the Internet, provided multicast routing is tuned.
795 
796 
797    I have no idea was this bicycle invented before me,
798    so that I had to set ARPHRD_IPGRE to a random value.
799    I have an impression, that Cisco could make something similar,
800    but this feature is apparently missing in IOS<=11.2(8).
801 
802    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
803    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
804 
805    ping -t 255 224.66.66.66
806 
807    If nobody answers, mbone does not work.
808 
809    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
810    ip addr add 10.66.66.<somewhat>/24 dev Universe
811    ifconfig Universe up
812    ifconfig Universe add fe80::<Your_real_addr>/10
813    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
814    ftp 10.66.66.66
815    ...
816    ftp fec0:6666:6666::193.233.7.65
817    ...
818  */
819 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
820 			unsigned short type,
821 			const void *daddr, const void *saddr, unsigned int len)
822 {
823 	struct ip_tunnel *t = netdev_priv(dev);
824 	struct iphdr *iph;
825 	struct gre_base_hdr *greh;
826 
827 	iph = skb_push(skb, t->hlen + sizeof(*iph));
828 	greh = (struct gre_base_hdr *)(iph+1);
829 	greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
830 	greh->protocol = htons(type);
831 
832 	memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
833 
834 	/* Set the source hardware address. */
835 	if (saddr)
836 		memcpy(&iph->saddr, saddr, 4);
837 	if (daddr)
838 		memcpy(&iph->daddr, daddr, 4);
839 	if (iph->daddr)
840 		return t->hlen + sizeof(*iph);
841 
842 	return -(t->hlen + sizeof(*iph));
843 }
844 
845 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
846 {
847 	const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
848 	memcpy(haddr, &iph->saddr, 4);
849 	return 4;
850 }
851 
852 static const struct header_ops ipgre_header_ops = {
853 	.create	= ipgre_header,
854 	.parse	= ipgre_header_parse,
855 };
856 
857 #ifdef CONFIG_NET_IPGRE_BROADCAST
858 static int ipgre_open(struct net_device *dev)
859 {
860 	struct ip_tunnel *t = netdev_priv(dev);
861 
862 	if (ipv4_is_multicast(t->parms.iph.daddr)) {
863 		struct flowi4 fl4;
864 		struct rtable *rt;
865 
866 		rt = ip_route_output_gre(t->net, &fl4,
867 					 t->parms.iph.daddr,
868 					 t->parms.iph.saddr,
869 					 t->parms.o_key,
870 					 RT_TOS(t->parms.iph.tos),
871 					 t->parms.link);
872 		if (IS_ERR(rt))
873 			return -EADDRNOTAVAIL;
874 		dev = rt->dst.dev;
875 		ip_rt_put(rt);
876 		if (!__in_dev_get_rtnl(dev))
877 			return -EADDRNOTAVAIL;
878 		t->mlink = dev->ifindex;
879 		ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
880 	}
881 	return 0;
882 }
883 
884 static int ipgre_close(struct net_device *dev)
885 {
886 	struct ip_tunnel *t = netdev_priv(dev);
887 
888 	if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
889 		struct in_device *in_dev;
890 		in_dev = inetdev_by_index(t->net, t->mlink);
891 		if (in_dev)
892 			ip_mc_dec_group(in_dev, t->parms.iph.daddr);
893 	}
894 	return 0;
895 }
896 #endif
897 
898 static const struct net_device_ops ipgre_netdev_ops = {
899 	.ndo_init		= ipgre_tunnel_init,
900 	.ndo_uninit		= ip_tunnel_uninit,
901 #ifdef CONFIG_NET_IPGRE_BROADCAST
902 	.ndo_open		= ipgre_open,
903 	.ndo_stop		= ipgre_close,
904 #endif
905 	.ndo_start_xmit		= ipgre_xmit,
906 	.ndo_do_ioctl		= ipgre_tunnel_ioctl,
907 	.ndo_change_mtu		= ip_tunnel_change_mtu,
908 	.ndo_get_stats64	= ip_tunnel_get_stats64,
909 	.ndo_get_iflink		= ip_tunnel_get_iflink,
910 };
911 
912 #define GRE_FEATURES (NETIF_F_SG |		\
913 		      NETIF_F_FRAGLIST |	\
914 		      NETIF_F_HIGHDMA |		\
915 		      NETIF_F_HW_CSUM)
916 
917 static void ipgre_tunnel_setup(struct net_device *dev)
918 {
919 	dev->netdev_ops		= &ipgre_netdev_ops;
920 	dev->type		= ARPHRD_IPGRE;
921 	ip_tunnel_setup(dev, ipgre_net_id);
922 }
923 
924 static void __gre_tunnel_init(struct net_device *dev)
925 {
926 	struct ip_tunnel *tunnel;
927 
928 	tunnel = netdev_priv(dev);
929 	tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
930 	tunnel->parms.iph.protocol = IPPROTO_GRE;
931 
932 	tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
933 
934 	dev->features		|= GRE_FEATURES;
935 	dev->hw_features	|= GRE_FEATURES;
936 
937 	if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
938 		/* TCP offload with GRE SEQ is not supported, nor
939 		 * can we support 2 levels of outer headers requiring
940 		 * an update.
941 		 */
942 		if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
943 		    (tunnel->encap.type == TUNNEL_ENCAP_NONE)) {
944 			dev->features    |= NETIF_F_GSO_SOFTWARE;
945 			dev->hw_features |= NETIF_F_GSO_SOFTWARE;
946 		}
947 
948 		/* Can use a lockless transmit, unless we generate
949 		 * output sequences
950 		 */
951 		dev->features |= NETIF_F_LLTX;
952 	}
953 }
954 
955 static int ipgre_tunnel_init(struct net_device *dev)
956 {
957 	struct ip_tunnel *tunnel = netdev_priv(dev);
958 	struct iphdr *iph = &tunnel->parms.iph;
959 
960 	__gre_tunnel_init(dev);
961 
962 	memcpy(dev->dev_addr, &iph->saddr, 4);
963 	memcpy(dev->broadcast, &iph->daddr, 4);
964 
965 	dev->flags		= IFF_NOARP;
966 	netif_keep_dst(dev);
967 	dev->addr_len		= 4;
968 
969 	if (iph->daddr && !tunnel->collect_md) {
970 #ifdef CONFIG_NET_IPGRE_BROADCAST
971 		if (ipv4_is_multicast(iph->daddr)) {
972 			if (!iph->saddr)
973 				return -EINVAL;
974 			dev->flags = IFF_BROADCAST;
975 			dev->header_ops = &ipgre_header_ops;
976 		}
977 #endif
978 	} else if (!tunnel->collect_md) {
979 		dev->header_ops = &ipgre_header_ops;
980 	}
981 
982 	return ip_tunnel_init(dev);
983 }
984 
985 static const struct gre_protocol ipgre_protocol = {
986 	.handler     = gre_rcv,
987 	.err_handler = gre_err,
988 };
989 
990 static int __net_init ipgre_init_net(struct net *net)
991 {
992 	return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
993 }
994 
995 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net)
996 {
997 	ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops);
998 }
999 
1000 static struct pernet_operations ipgre_net_ops = {
1001 	.init = ipgre_init_net,
1002 	.exit_batch = ipgre_exit_batch_net,
1003 	.id   = &ipgre_net_id,
1004 	.size = sizeof(struct ip_tunnel_net),
1005 };
1006 
1007 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1008 				 struct netlink_ext_ack *extack)
1009 {
1010 	__be16 flags;
1011 
1012 	if (!data)
1013 		return 0;
1014 
1015 	flags = 0;
1016 	if (data[IFLA_GRE_IFLAGS])
1017 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1018 	if (data[IFLA_GRE_OFLAGS])
1019 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1020 	if (flags & (GRE_VERSION|GRE_ROUTING))
1021 		return -EINVAL;
1022 
1023 	if (data[IFLA_GRE_COLLECT_METADATA] &&
1024 	    data[IFLA_GRE_ENCAP_TYPE] &&
1025 	    nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1026 		return -EINVAL;
1027 
1028 	return 0;
1029 }
1030 
1031 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1032 			      struct netlink_ext_ack *extack)
1033 {
1034 	__be32 daddr;
1035 
1036 	if (tb[IFLA_ADDRESS]) {
1037 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1038 			return -EINVAL;
1039 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1040 			return -EADDRNOTAVAIL;
1041 	}
1042 
1043 	if (!data)
1044 		goto out;
1045 
1046 	if (data[IFLA_GRE_REMOTE]) {
1047 		memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1048 		if (!daddr)
1049 			return -EINVAL;
1050 	}
1051 
1052 out:
1053 	return ipgre_tunnel_validate(tb, data, extack);
1054 }
1055 
1056 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1057 			   struct netlink_ext_ack *extack)
1058 {
1059 	__be16 flags = 0;
1060 	int ret;
1061 
1062 	if (!data)
1063 		return 0;
1064 
1065 	ret = ipgre_tap_validate(tb, data, extack);
1066 	if (ret)
1067 		return ret;
1068 
1069 	/* ERSPAN should only have GRE sequence and key flag */
1070 	if (data[IFLA_GRE_OFLAGS])
1071 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1072 	if (data[IFLA_GRE_IFLAGS])
1073 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1074 	if (!data[IFLA_GRE_COLLECT_METADATA] &&
1075 	    flags != (GRE_SEQ | GRE_KEY))
1076 		return -EINVAL;
1077 
1078 	/* ERSPAN Session ID only has 10-bit. Since we reuse
1079 	 * 32-bit key field as ID, check it's range.
1080 	 */
1081 	if (data[IFLA_GRE_IKEY] &&
1082 	    (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1083 		return -EINVAL;
1084 
1085 	if (data[IFLA_GRE_OKEY] &&
1086 	    (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1087 		return -EINVAL;
1088 
1089 	return 0;
1090 }
1091 
1092 static int ipgre_netlink_parms(struct net_device *dev,
1093 				struct nlattr *data[],
1094 				struct nlattr *tb[],
1095 				struct ip_tunnel_parm *parms,
1096 				__u32 *fwmark)
1097 {
1098 	struct ip_tunnel *t = netdev_priv(dev);
1099 
1100 	memset(parms, 0, sizeof(*parms));
1101 
1102 	parms->iph.protocol = IPPROTO_GRE;
1103 
1104 	if (!data)
1105 		return 0;
1106 
1107 	if (data[IFLA_GRE_LINK])
1108 		parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1109 
1110 	if (data[IFLA_GRE_IFLAGS])
1111 		parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
1112 
1113 	if (data[IFLA_GRE_OFLAGS])
1114 		parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
1115 
1116 	if (data[IFLA_GRE_IKEY])
1117 		parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1118 
1119 	if (data[IFLA_GRE_OKEY])
1120 		parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1121 
1122 	if (data[IFLA_GRE_LOCAL])
1123 		parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1124 
1125 	if (data[IFLA_GRE_REMOTE])
1126 		parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1127 
1128 	if (data[IFLA_GRE_TTL])
1129 		parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1130 
1131 	if (data[IFLA_GRE_TOS])
1132 		parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1133 
1134 	if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1135 		if (t->ignore_df)
1136 			return -EINVAL;
1137 		parms->iph.frag_off = htons(IP_DF);
1138 	}
1139 
1140 	if (data[IFLA_GRE_COLLECT_METADATA]) {
1141 		t->collect_md = true;
1142 		if (dev->type == ARPHRD_IPGRE)
1143 			dev->type = ARPHRD_NONE;
1144 	}
1145 
1146 	if (data[IFLA_GRE_IGNORE_DF]) {
1147 		if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1148 		  && (parms->iph.frag_off & htons(IP_DF)))
1149 			return -EINVAL;
1150 		t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1151 	}
1152 
1153 	if (data[IFLA_GRE_FWMARK])
1154 		*fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1155 
1156 	if (data[IFLA_GRE_ERSPAN_VER]) {
1157 		t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]);
1158 
1159 		if (t->erspan_ver != 1 && t->erspan_ver != 2)
1160 			return -EINVAL;
1161 	}
1162 
1163 	if (t->erspan_ver == 1) {
1164 		if (data[IFLA_GRE_ERSPAN_INDEX]) {
1165 			t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1166 			if (t->index & ~INDEX_MASK)
1167 				return -EINVAL;
1168 		}
1169 	} else if (t->erspan_ver == 2) {
1170 		if (data[IFLA_GRE_ERSPAN_DIR]) {
1171 			t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]);
1172 			if (t->dir & ~(DIR_MASK >> DIR_OFFSET))
1173 				return -EINVAL;
1174 		}
1175 		if (data[IFLA_GRE_ERSPAN_HWID]) {
1176 			t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]);
1177 			if (t->hwid & ~(HWID_MASK >> HWID_OFFSET))
1178 				return -EINVAL;
1179 		}
1180 	}
1181 
1182 	return 0;
1183 }
1184 
1185 /* This function returns true when ENCAP attributes are present in the nl msg */
1186 static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1187 				      struct ip_tunnel_encap *ipencap)
1188 {
1189 	bool ret = false;
1190 
1191 	memset(ipencap, 0, sizeof(*ipencap));
1192 
1193 	if (!data)
1194 		return ret;
1195 
1196 	if (data[IFLA_GRE_ENCAP_TYPE]) {
1197 		ret = true;
1198 		ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1199 	}
1200 
1201 	if (data[IFLA_GRE_ENCAP_FLAGS]) {
1202 		ret = true;
1203 		ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1204 	}
1205 
1206 	if (data[IFLA_GRE_ENCAP_SPORT]) {
1207 		ret = true;
1208 		ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1209 	}
1210 
1211 	if (data[IFLA_GRE_ENCAP_DPORT]) {
1212 		ret = true;
1213 		ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1214 	}
1215 
1216 	return ret;
1217 }
1218 
1219 static int gre_tap_init(struct net_device *dev)
1220 {
1221 	__gre_tunnel_init(dev);
1222 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1223 	netif_keep_dst(dev);
1224 
1225 	return ip_tunnel_init(dev);
1226 }
1227 
1228 static const struct net_device_ops gre_tap_netdev_ops = {
1229 	.ndo_init		= gre_tap_init,
1230 	.ndo_uninit		= ip_tunnel_uninit,
1231 	.ndo_start_xmit		= gre_tap_xmit,
1232 	.ndo_set_mac_address 	= eth_mac_addr,
1233 	.ndo_validate_addr	= eth_validate_addr,
1234 	.ndo_change_mtu		= ip_tunnel_change_mtu,
1235 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1236 	.ndo_get_iflink		= ip_tunnel_get_iflink,
1237 	.ndo_fill_metadata_dst	= gre_fill_metadata_dst,
1238 };
1239 
1240 static int erspan_tunnel_init(struct net_device *dev)
1241 {
1242 	struct ip_tunnel *tunnel = netdev_priv(dev);
1243 
1244 	tunnel->tun_hlen = 8;
1245 	tunnel->parms.iph.protocol = IPPROTO_GRE;
1246 	tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1247 		       erspan_hdr_len(tunnel->erspan_ver);
1248 
1249 	dev->features		|= GRE_FEATURES;
1250 	dev->hw_features	|= GRE_FEATURES;
1251 	dev->priv_flags		|= IFF_LIVE_ADDR_CHANGE;
1252 	netif_keep_dst(dev);
1253 
1254 	return ip_tunnel_init(dev);
1255 }
1256 
1257 static const struct net_device_ops erspan_netdev_ops = {
1258 	.ndo_init		= erspan_tunnel_init,
1259 	.ndo_uninit		= ip_tunnel_uninit,
1260 	.ndo_start_xmit		= erspan_xmit,
1261 	.ndo_set_mac_address	= eth_mac_addr,
1262 	.ndo_validate_addr	= eth_validate_addr,
1263 	.ndo_change_mtu		= ip_tunnel_change_mtu,
1264 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1265 	.ndo_get_iflink		= ip_tunnel_get_iflink,
1266 	.ndo_fill_metadata_dst	= gre_fill_metadata_dst,
1267 };
1268 
1269 static void ipgre_tap_setup(struct net_device *dev)
1270 {
1271 	ether_setup(dev);
1272 	dev->max_mtu = 0;
1273 	dev->netdev_ops	= &gre_tap_netdev_ops;
1274 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1275 	dev->priv_flags	|= IFF_LIVE_ADDR_CHANGE;
1276 	ip_tunnel_setup(dev, gre_tap_net_id);
1277 }
1278 
1279 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1280 			 struct nlattr *tb[], struct nlattr *data[],
1281 			 struct netlink_ext_ack *extack)
1282 {
1283 	struct ip_tunnel_parm p;
1284 	struct ip_tunnel_encap ipencap;
1285 	__u32 fwmark = 0;
1286 	int err;
1287 
1288 	if (ipgre_netlink_encap_parms(data, &ipencap)) {
1289 		struct ip_tunnel *t = netdev_priv(dev);
1290 		err = ip_tunnel_encap_setup(t, &ipencap);
1291 
1292 		if (err < 0)
1293 			return err;
1294 	}
1295 
1296 	err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1297 	if (err < 0)
1298 		return err;
1299 	return ip_tunnel_newlink(dev, tb, &p, fwmark);
1300 }
1301 
1302 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1303 			    struct nlattr *data[],
1304 			    struct netlink_ext_ack *extack)
1305 {
1306 	struct ip_tunnel *t = netdev_priv(dev);
1307 	struct ip_tunnel_encap ipencap;
1308 	__u32 fwmark = t->fwmark;
1309 	struct ip_tunnel_parm p;
1310 	int err;
1311 
1312 	if (ipgre_netlink_encap_parms(data, &ipencap)) {
1313 		err = ip_tunnel_encap_setup(t, &ipencap);
1314 
1315 		if (err < 0)
1316 			return err;
1317 	}
1318 
1319 	err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1320 	if (err < 0)
1321 		return err;
1322 
1323 	err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1324 	if (err < 0)
1325 		return err;
1326 
1327 	t->parms.i_flags = p.i_flags;
1328 	t->parms.o_flags = p.o_flags;
1329 
1330 	if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
1331 		ipgre_link_update(dev, !tb[IFLA_MTU]);
1332 
1333 	return 0;
1334 }
1335 
1336 static size_t ipgre_get_size(const struct net_device *dev)
1337 {
1338 	return
1339 		/* IFLA_GRE_LINK */
1340 		nla_total_size(4) +
1341 		/* IFLA_GRE_IFLAGS */
1342 		nla_total_size(2) +
1343 		/* IFLA_GRE_OFLAGS */
1344 		nla_total_size(2) +
1345 		/* IFLA_GRE_IKEY */
1346 		nla_total_size(4) +
1347 		/* IFLA_GRE_OKEY */
1348 		nla_total_size(4) +
1349 		/* IFLA_GRE_LOCAL */
1350 		nla_total_size(4) +
1351 		/* IFLA_GRE_REMOTE */
1352 		nla_total_size(4) +
1353 		/* IFLA_GRE_TTL */
1354 		nla_total_size(1) +
1355 		/* IFLA_GRE_TOS */
1356 		nla_total_size(1) +
1357 		/* IFLA_GRE_PMTUDISC */
1358 		nla_total_size(1) +
1359 		/* IFLA_GRE_ENCAP_TYPE */
1360 		nla_total_size(2) +
1361 		/* IFLA_GRE_ENCAP_FLAGS */
1362 		nla_total_size(2) +
1363 		/* IFLA_GRE_ENCAP_SPORT */
1364 		nla_total_size(2) +
1365 		/* IFLA_GRE_ENCAP_DPORT */
1366 		nla_total_size(2) +
1367 		/* IFLA_GRE_COLLECT_METADATA */
1368 		nla_total_size(0) +
1369 		/* IFLA_GRE_IGNORE_DF */
1370 		nla_total_size(1) +
1371 		/* IFLA_GRE_FWMARK */
1372 		nla_total_size(4) +
1373 		/* IFLA_GRE_ERSPAN_INDEX */
1374 		nla_total_size(4) +
1375 		/* IFLA_GRE_ERSPAN_VER */
1376 		nla_total_size(1) +
1377 		/* IFLA_GRE_ERSPAN_DIR */
1378 		nla_total_size(1) +
1379 		/* IFLA_GRE_ERSPAN_HWID */
1380 		nla_total_size(2) +
1381 		0;
1382 }
1383 
1384 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1385 {
1386 	struct ip_tunnel *t = netdev_priv(dev);
1387 	struct ip_tunnel_parm *p = &t->parms;
1388 	__be16 o_flags = p->o_flags;
1389 
1390 	if (t->erspan_ver == 1 || t->erspan_ver == 2) {
1391 		if (!t->collect_md)
1392 			o_flags |= TUNNEL_KEY;
1393 
1394 		if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1395 			goto nla_put_failure;
1396 
1397 		if (t->erspan_ver == 1) {
1398 			if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1399 				goto nla_put_failure;
1400 		} else {
1401 			if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1402 				goto nla_put_failure;
1403 			if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1404 				goto nla_put_failure;
1405 		}
1406 	}
1407 
1408 	if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1409 	    nla_put_be16(skb, IFLA_GRE_IFLAGS,
1410 			 gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1411 	    nla_put_be16(skb, IFLA_GRE_OFLAGS,
1412 			 gre_tnl_flags_to_gre_flags(o_flags)) ||
1413 	    nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1414 	    nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1415 	    nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1416 	    nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1417 	    nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1418 	    nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1419 	    nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1420 		       !!(p->iph.frag_off & htons(IP_DF))) ||
1421 	    nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1422 		goto nla_put_failure;
1423 
1424 	if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1425 			t->encap.type) ||
1426 	    nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1427 			 t->encap.sport) ||
1428 	    nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1429 			 t->encap.dport) ||
1430 	    nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1431 			t->encap.flags))
1432 		goto nla_put_failure;
1433 
1434 	if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1435 		goto nla_put_failure;
1436 
1437 	if (t->collect_md) {
1438 		if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1439 			goto nla_put_failure;
1440 	}
1441 
1442 	return 0;
1443 
1444 nla_put_failure:
1445 	return -EMSGSIZE;
1446 }
1447 
1448 static void erspan_setup(struct net_device *dev)
1449 {
1450 	struct ip_tunnel *t = netdev_priv(dev);
1451 
1452 	ether_setup(dev);
1453 	dev->max_mtu = 0;
1454 	dev->netdev_ops = &erspan_netdev_ops;
1455 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1456 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1457 	ip_tunnel_setup(dev, erspan_net_id);
1458 	t->erspan_ver = 1;
1459 }
1460 
1461 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1462 	[IFLA_GRE_LINK]		= { .type = NLA_U32 },
1463 	[IFLA_GRE_IFLAGS]	= { .type = NLA_U16 },
1464 	[IFLA_GRE_OFLAGS]	= { .type = NLA_U16 },
1465 	[IFLA_GRE_IKEY]		= { .type = NLA_U32 },
1466 	[IFLA_GRE_OKEY]		= { .type = NLA_U32 },
1467 	[IFLA_GRE_LOCAL]	= { .len = sizeof_field(struct iphdr, saddr) },
1468 	[IFLA_GRE_REMOTE]	= { .len = sizeof_field(struct iphdr, daddr) },
1469 	[IFLA_GRE_TTL]		= { .type = NLA_U8 },
1470 	[IFLA_GRE_TOS]		= { .type = NLA_U8 },
1471 	[IFLA_GRE_PMTUDISC]	= { .type = NLA_U8 },
1472 	[IFLA_GRE_ENCAP_TYPE]	= { .type = NLA_U16 },
1473 	[IFLA_GRE_ENCAP_FLAGS]	= { .type = NLA_U16 },
1474 	[IFLA_GRE_ENCAP_SPORT]	= { .type = NLA_U16 },
1475 	[IFLA_GRE_ENCAP_DPORT]	= { .type = NLA_U16 },
1476 	[IFLA_GRE_COLLECT_METADATA]	= { .type = NLA_FLAG },
1477 	[IFLA_GRE_IGNORE_DF]	= { .type = NLA_U8 },
1478 	[IFLA_GRE_FWMARK]	= { .type = NLA_U32 },
1479 	[IFLA_GRE_ERSPAN_INDEX]	= { .type = NLA_U32 },
1480 	[IFLA_GRE_ERSPAN_VER]	= { .type = NLA_U8 },
1481 	[IFLA_GRE_ERSPAN_DIR]	= { .type = NLA_U8 },
1482 	[IFLA_GRE_ERSPAN_HWID]	= { .type = NLA_U16 },
1483 };
1484 
1485 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1486 	.kind		= "gre",
1487 	.maxtype	= IFLA_GRE_MAX,
1488 	.policy		= ipgre_policy,
1489 	.priv_size	= sizeof(struct ip_tunnel),
1490 	.setup		= ipgre_tunnel_setup,
1491 	.validate	= ipgre_tunnel_validate,
1492 	.newlink	= ipgre_newlink,
1493 	.changelink	= ipgre_changelink,
1494 	.dellink	= ip_tunnel_dellink,
1495 	.get_size	= ipgre_get_size,
1496 	.fill_info	= ipgre_fill_info,
1497 	.get_link_net	= ip_tunnel_get_link_net,
1498 };
1499 
1500 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1501 	.kind		= "gretap",
1502 	.maxtype	= IFLA_GRE_MAX,
1503 	.policy		= ipgre_policy,
1504 	.priv_size	= sizeof(struct ip_tunnel),
1505 	.setup		= ipgre_tap_setup,
1506 	.validate	= ipgre_tap_validate,
1507 	.newlink	= ipgre_newlink,
1508 	.changelink	= ipgre_changelink,
1509 	.dellink	= ip_tunnel_dellink,
1510 	.get_size	= ipgre_get_size,
1511 	.fill_info	= ipgre_fill_info,
1512 	.get_link_net	= ip_tunnel_get_link_net,
1513 };
1514 
1515 static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1516 	.kind		= "erspan",
1517 	.maxtype	= IFLA_GRE_MAX,
1518 	.policy		= ipgre_policy,
1519 	.priv_size	= sizeof(struct ip_tunnel),
1520 	.setup		= erspan_setup,
1521 	.validate	= erspan_validate,
1522 	.newlink	= ipgre_newlink,
1523 	.changelink	= ipgre_changelink,
1524 	.dellink	= ip_tunnel_dellink,
1525 	.get_size	= ipgre_get_size,
1526 	.fill_info	= ipgre_fill_info,
1527 	.get_link_net	= ip_tunnel_get_link_net,
1528 };
1529 
1530 struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1531 					u8 name_assign_type)
1532 {
1533 	struct nlattr *tb[IFLA_MAX + 1];
1534 	struct net_device *dev;
1535 	LIST_HEAD(list_kill);
1536 	struct ip_tunnel *t;
1537 	int err;
1538 
1539 	memset(&tb, 0, sizeof(tb));
1540 
1541 	dev = rtnl_create_link(net, name, name_assign_type,
1542 			       &ipgre_tap_ops, tb, NULL);
1543 	if (IS_ERR(dev))
1544 		return dev;
1545 
1546 	/* Configure flow based GRE device. */
1547 	t = netdev_priv(dev);
1548 	t->collect_md = true;
1549 
1550 	err = ipgre_newlink(net, dev, tb, NULL, NULL);
1551 	if (err < 0) {
1552 		free_netdev(dev);
1553 		return ERR_PTR(err);
1554 	}
1555 
1556 	/* openvswitch users expect packet sizes to be unrestricted,
1557 	 * so set the largest MTU we can.
1558 	 */
1559 	err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1560 	if (err)
1561 		goto out;
1562 
1563 	err = rtnl_configure_link(dev, NULL);
1564 	if (err < 0)
1565 		goto out;
1566 
1567 	return dev;
1568 out:
1569 	ip_tunnel_dellink(dev, &list_kill);
1570 	unregister_netdevice_many(&list_kill);
1571 	return ERR_PTR(err);
1572 }
1573 EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1574 
1575 static int __net_init ipgre_tap_init_net(struct net *net)
1576 {
1577 	return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1578 }
1579 
1580 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net)
1581 {
1582 	ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops);
1583 }
1584 
1585 static struct pernet_operations ipgre_tap_net_ops = {
1586 	.init = ipgre_tap_init_net,
1587 	.exit_batch = ipgre_tap_exit_batch_net,
1588 	.id   = &gre_tap_net_id,
1589 	.size = sizeof(struct ip_tunnel_net),
1590 };
1591 
1592 static int __net_init erspan_init_net(struct net *net)
1593 {
1594 	return ip_tunnel_init_net(net, erspan_net_id,
1595 				  &erspan_link_ops, "erspan0");
1596 }
1597 
1598 static void __net_exit erspan_exit_batch_net(struct list_head *net_list)
1599 {
1600 	ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops);
1601 }
1602 
1603 static struct pernet_operations erspan_net_ops = {
1604 	.init = erspan_init_net,
1605 	.exit_batch = erspan_exit_batch_net,
1606 	.id   = &erspan_net_id,
1607 	.size = sizeof(struct ip_tunnel_net),
1608 };
1609 
1610 static int __init ipgre_init(void)
1611 {
1612 	int err;
1613 
1614 	pr_info("GRE over IPv4 tunneling driver\n");
1615 
1616 	err = register_pernet_device(&ipgre_net_ops);
1617 	if (err < 0)
1618 		return err;
1619 
1620 	err = register_pernet_device(&ipgre_tap_net_ops);
1621 	if (err < 0)
1622 		goto pnet_tap_failed;
1623 
1624 	err = register_pernet_device(&erspan_net_ops);
1625 	if (err < 0)
1626 		goto pnet_erspan_failed;
1627 
1628 	err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1629 	if (err < 0) {
1630 		pr_info("%s: can't add protocol\n", __func__);
1631 		goto add_proto_failed;
1632 	}
1633 
1634 	err = rtnl_link_register(&ipgre_link_ops);
1635 	if (err < 0)
1636 		goto rtnl_link_failed;
1637 
1638 	err = rtnl_link_register(&ipgre_tap_ops);
1639 	if (err < 0)
1640 		goto tap_ops_failed;
1641 
1642 	err = rtnl_link_register(&erspan_link_ops);
1643 	if (err < 0)
1644 		goto erspan_link_failed;
1645 
1646 	return 0;
1647 
1648 erspan_link_failed:
1649 	rtnl_link_unregister(&ipgre_tap_ops);
1650 tap_ops_failed:
1651 	rtnl_link_unregister(&ipgre_link_ops);
1652 rtnl_link_failed:
1653 	gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1654 add_proto_failed:
1655 	unregister_pernet_device(&erspan_net_ops);
1656 pnet_erspan_failed:
1657 	unregister_pernet_device(&ipgre_tap_net_ops);
1658 pnet_tap_failed:
1659 	unregister_pernet_device(&ipgre_net_ops);
1660 	return err;
1661 }
1662 
1663 static void __exit ipgre_fini(void)
1664 {
1665 	rtnl_link_unregister(&ipgre_tap_ops);
1666 	rtnl_link_unregister(&ipgre_link_ops);
1667 	rtnl_link_unregister(&erspan_link_ops);
1668 	gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1669 	unregister_pernet_device(&ipgre_tap_net_ops);
1670 	unregister_pernet_device(&ipgre_net_ops);
1671 	unregister_pernet_device(&erspan_net_ops);
1672 }
1673 
1674 module_init(ipgre_init);
1675 module_exit(ipgre_fini);
1676 MODULE_LICENSE("GPL");
1677 MODULE_ALIAS_RTNL_LINK("gre");
1678 MODULE_ALIAS_RTNL_LINK("gretap");
1679 MODULE_ALIAS_RTNL_LINK("erspan");
1680 MODULE_ALIAS_NETDEV("gre0");
1681 MODULE_ALIAS_NETDEV("gretap0");
1682 MODULE_ALIAS_NETDEV("erspan0");
1683