xref: /openbmc/linux/net/ipv4/ip_gre.c (revision 9dae47aba0a055f761176d9297371d5bb24289ec)
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 				__be32 id, u32 index,
118 				bool truncate, bool is_ipv4);
119 
120 static unsigned int ipgre_net_id __read_mostly;
121 static unsigned int gre_tap_net_id __read_mostly;
122 static unsigned int erspan_net_id __read_mostly;
123 
124 static void ipgre_err(struct sk_buff *skb, u32 info,
125 		      const struct tnl_ptk_info *tpi)
126 {
127 
128 	/* All the routers (except for Linux) return only
129 	   8 bytes of packet payload. It means, that precise relaying of
130 	   ICMP in the real Internet is absolutely infeasible.
131 
132 	   Moreover, Cisco "wise men" put GRE key to the third word
133 	   in GRE header. It makes impossible maintaining even soft
134 	   state for keyed GRE tunnels with enabled checksum. Tell
135 	   them "thank you".
136 
137 	   Well, I wonder, rfc1812 was written by Cisco employee,
138 	   what the hell these idiots break standards established
139 	   by themselves???
140 	   */
141 	struct net *net = dev_net(skb->dev);
142 	struct ip_tunnel_net *itn;
143 	const struct iphdr *iph;
144 	const int type = icmp_hdr(skb)->type;
145 	const int code = icmp_hdr(skb)->code;
146 	unsigned int data_len = 0;
147 	struct ip_tunnel *t;
148 
149 	switch (type) {
150 	default:
151 	case ICMP_PARAMETERPROB:
152 		return;
153 
154 	case ICMP_DEST_UNREACH:
155 		switch (code) {
156 		case ICMP_SR_FAILED:
157 		case ICMP_PORT_UNREACH:
158 			/* Impossible event. */
159 			return;
160 		default:
161 			/* All others are translated to HOST_UNREACH.
162 			   rfc2003 contains "deep thoughts" about NET_UNREACH,
163 			   I believe they are just ether pollution. --ANK
164 			 */
165 			break;
166 		}
167 		break;
168 
169 	case ICMP_TIME_EXCEEDED:
170 		if (code != ICMP_EXC_TTL)
171 			return;
172 		data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */
173 		break;
174 
175 	case ICMP_REDIRECT:
176 		break;
177 	}
178 
179 	if (tpi->proto == htons(ETH_P_TEB))
180 		itn = net_generic(net, gre_tap_net_id);
181 	else
182 		itn = net_generic(net, ipgre_net_id);
183 
184 	iph = (const struct iphdr *)(icmp_hdr(skb) + 1);
185 	t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
186 			     iph->daddr, iph->saddr, tpi->key);
187 
188 	if (!t)
189 		return;
190 
191 #if IS_ENABLED(CONFIG_IPV6)
192        if (tpi->proto == htons(ETH_P_IPV6) &&
193            !ip6_err_gen_icmpv6_unreach(skb, iph->ihl * 4 + tpi->hdr_len,
194 				       type, data_len))
195                return;
196 #endif
197 
198 	if (t->parms.iph.daddr == 0 ||
199 	    ipv4_is_multicast(t->parms.iph.daddr))
200 		return;
201 
202 	if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
203 		return;
204 
205 	if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
206 		t->err_count++;
207 	else
208 		t->err_count = 1;
209 	t->err_time = jiffies;
210 }
211 
212 static void gre_err(struct sk_buff *skb, u32 info)
213 {
214 	/* All the routers (except for Linux) return only
215 	 * 8 bytes of packet payload. It means, that precise relaying of
216 	 * ICMP in the real Internet is absolutely infeasible.
217 	 *
218 	 * Moreover, Cisco "wise men" put GRE key to the third word
219 	 * in GRE header. It makes impossible maintaining even soft
220 	 * state for keyed
221 	 * GRE tunnels with enabled checksum. Tell them "thank you".
222 	 *
223 	 * Well, I wonder, rfc1812 was written by Cisco employee,
224 	 * what the hell these idiots break standards established
225 	 * by themselves???
226 	 */
227 
228 	const struct iphdr *iph = (struct iphdr *)skb->data;
229 	const int type = icmp_hdr(skb)->type;
230 	const int code = icmp_hdr(skb)->code;
231 	struct tnl_ptk_info tpi;
232 	bool csum_err = false;
233 
234 	if (gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP),
235 			     iph->ihl * 4) < 0) {
236 		if (!csum_err)		/* ignore csum errors. */
237 			return;
238 	}
239 
240 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
241 		ipv4_update_pmtu(skb, dev_net(skb->dev), info,
242 				 skb->dev->ifindex, 0, IPPROTO_GRE, 0);
243 		return;
244 	}
245 	if (type == ICMP_REDIRECT) {
246 		ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex, 0,
247 			      IPPROTO_GRE, 0);
248 		return;
249 	}
250 
251 	ipgre_err(skb, info, &tpi);
252 }
253 
254 static int erspan_rcv(struct sk_buff *skb, struct tnl_ptk_info *tpi,
255 		      int gre_hdr_len)
256 {
257 	struct net *net = dev_net(skb->dev);
258 	struct metadata_dst *tun_dst = NULL;
259 	struct erspan_base_hdr *ershdr;
260 	struct erspan_metadata *pkt_md;
261 	struct ip_tunnel_net *itn;
262 	struct ip_tunnel *tunnel;
263 	const struct iphdr *iph;
264 	int ver;
265 	int len;
266 
267 	itn = net_generic(net, erspan_net_id);
268 	len = gre_hdr_len + sizeof(*ershdr);
269 
270 	/* Check based hdr len */
271 	if (unlikely(!pskb_may_pull(skb, len)))
272 		return PACKET_REJECT;
273 
274 	iph = ip_hdr(skb);
275 	ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
276 	ver = (ntohs(ershdr->ver_vlan) & VER_MASK) >> VER_OFFSET;
277 
278 	/* The original GRE header does not have key field,
279 	 * Use ERSPAN 10-bit session ID as key.
280 	 */
281 	tpi->key = cpu_to_be32(ntohs(ershdr->session_id) & ID_MASK);
282 	tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex,
283 				  tpi->flags | TUNNEL_KEY,
284 				  iph->saddr, iph->daddr, tpi->key);
285 
286 	if (tunnel) {
287 		len = gre_hdr_len + erspan_hdr_len(ver);
288 		if (unlikely(!pskb_may_pull(skb, len)))
289 			return PACKET_REJECT;
290 
291 		ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
292 		pkt_md = (struct erspan_metadata *)(ershdr + 1);
293 
294 		if (__iptunnel_pull_header(skb,
295 					   len,
296 					   htons(ETH_P_TEB),
297 					   false, false) < 0)
298 			goto drop;
299 
300 		if (tunnel->collect_md) {
301 			struct ip_tunnel_info *info;
302 			struct erspan_metadata *md;
303 			__be64 tun_id;
304 			__be16 flags;
305 
306 			tpi->flags |= TUNNEL_KEY;
307 			flags = tpi->flags;
308 			tun_id = key32_to_tunnel_id(tpi->key);
309 
310 			tun_dst = ip_tun_rx_dst(skb, flags,
311 						tun_id, sizeof(*md));
312 			if (!tun_dst)
313 				return PACKET_REJECT;
314 
315 			md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
316 			memcpy(md, pkt_md, sizeof(*md));
317 			md->version = ver;
318 
319 			info = &tun_dst->u.tun_info;
320 			info->key.tun_flags |= TUNNEL_ERSPAN_OPT;
321 			info->options_len = sizeof(*md);
322 		} else {
323 			tunnel->erspan_ver = ver;
324 			if (ver == 1) {
325 				tunnel->index = ntohl(pkt_md->u.index);
326 			} else {
327 				u16 md2_flags;
328 				u16 dir, hwid;
329 
330 				md2_flags = ntohs(pkt_md->u.md2.flags);
331 				dir = (md2_flags & DIR_MASK) >> DIR_OFFSET;
332 				hwid = (md2_flags & HWID_MASK) >> HWID_OFFSET;
333 				tunnel->dir = dir;
334 				tunnel->hwid = hwid;
335 			}
336 
337 		}
338 
339 		skb_reset_mac_header(skb);
340 		ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
341 		return PACKET_RCVD;
342 	}
343 drop:
344 	kfree_skb(skb);
345 	return PACKET_RCVD;
346 }
347 
348 static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
349 		       struct ip_tunnel_net *itn, int hdr_len, bool raw_proto)
350 {
351 	struct metadata_dst *tun_dst = NULL;
352 	const struct iphdr *iph;
353 	struct ip_tunnel *tunnel;
354 
355 	iph = ip_hdr(skb);
356 	tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
357 				  iph->saddr, iph->daddr, tpi->key);
358 
359 	if (tunnel) {
360 		if (__iptunnel_pull_header(skb, hdr_len, tpi->proto,
361 					   raw_proto, false) < 0)
362 			goto drop;
363 
364 		if (tunnel->dev->type != ARPHRD_NONE)
365 			skb_pop_mac_header(skb);
366 		else
367 			skb_reset_mac_header(skb);
368 		if (tunnel->collect_md) {
369 			__be16 flags;
370 			__be64 tun_id;
371 
372 			flags = tpi->flags & (TUNNEL_CSUM | TUNNEL_KEY);
373 			tun_id = key32_to_tunnel_id(tpi->key);
374 			tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
375 			if (!tun_dst)
376 				return PACKET_REJECT;
377 		}
378 
379 		ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
380 		return PACKET_RCVD;
381 	}
382 	return PACKET_NEXT;
383 
384 drop:
385 	kfree_skb(skb);
386 	return PACKET_RCVD;
387 }
388 
389 static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
390 		     int hdr_len)
391 {
392 	struct net *net = dev_net(skb->dev);
393 	struct ip_tunnel_net *itn;
394 	int res;
395 
396 	if (tpi->proto == htons(ETH_P_TEB))
397 		itn = net_generic(net, gre_tap_net_id);
398 	else
399 		itn = net_generic(net, ipgre_net_id);
400 
401 	res = __ipgre_rcv(skb, tpi, itn, hdr_len, false);
402 	if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) {
403 		/* ipgre tunnels in collect metadata mode should receive
404 		 * also ETH_P_TEB traffic.
405 		 */
406 		itn = net_generic(net, ipgre_net_id);
407 		res = __ipgre_rcv(skb, tpi, itn, hdr_len, true);
408 	}
409 	return res;
410 }
411 
412 static int gre_rcv(struct sk_buff *skb)
413 {
414 	struct tnl_ptk_info tpi;
415 	bool csum_err = false;
416 	int hdr_len;
417 
418 #ifdef CONFIG_NET_IPGRE_BROADCAST
419 	if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
420 		/* Looped back packet, drop it! */
421 		if (rt_is_output_route(skb_rtable(skb)))
422 			goto drop;
423 	}
424 #endif
425 
426 	hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0);
427 	if (hdr_len < 0)
428 		goto drop;
429 
430 	if (unlikely(tpi.proto == htons(ETH_P_ERSPAN) ||
431 		     tpi.proto == htons(ETH_P_ERSPAN2))) {
432 		if (erspan_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
433 			return 0;
434 		goto out;
435 	}
436 
437 	if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
438 		return 0;
439 
440 out:
441 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
442 drop:
443 	kfree_skb(skb);
444 	return 0;
445 }
446 
447 static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
448 		       const struct iphdr *tnl_params,
449 		       __be16 proto)
450 {
451 	struct ip_tunnel *tunnel = netdev_priv(dev);
452 
453 	if (tunnel->parms.o_flags & TUNNEL_SEQ)
454 		tunnel->o_seqno++;
455 
456 	/* Push GRE header. */
457 	gre_build_header(skb, tunnel->tun_hlen,
458 			 tunnel->parms.o_flags, proto, tunnel->parms.o_key,
459 			 htonl(tunnel->o_seqno));
460 
461 	ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
462 }
463 
464 static int gre_handle_offloads(struct sk_buff *skb, bool csum)
465 {
466 	return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
467 }
468 
469 static struct rtable *gre_get_rt(struct sk_buff *skb,
470 				 struct net_device *dev,
471 				 struct flowi4 *fl,
472 				 const struct ip_tunnel_key *key)
473 {
474 	struct net *net = dev_net(dev);
475 
476 	memset(fl, 0, sizeof(*fl));
477 	fl->daddr = key->u.ipv4.dst;
478 	fl->saddr = key->u.ipv4.src;
479 	fl->flowi4_tos = RT_TOS(key->tos);
480 	fl->flowi4_mark = skb->mark;
481 	fl->flowi4_proto = IPPROTO_GRE;
482 
483 	return ip_route_output_key(net, fl);
484 }
485 
486 static struct rtable *prepare_fb_xmit(struct sk_buff *skb,
487 				      struct net_device *dev,
488 				      struct flowi4 *fl,
489 				      int tunnel_hlen)
490 {
491 	struct ip_tunnel_info *tun_info;
492 	const struct ip_tunnel_key *key;
493 	struct rtable *rt = NULL;
494 	int min_headroom;
495 	bool use_cache;
496 	int err;
497 
498 	tun_info = skb_tunnel_info(skb);
499 	key = &tun_info->key;
500 	use_cache = ip_tunnel_dst_cache_usable(skb, tun_info);
501 
502 	if (use_cache)
503 		rt = dst_cache_get_ip4(&tun_info->dst_cache, &fl->saddr);
504 	if (!rt) {
505 		rt = gre_get_rt(skb, dev, fl, key);
506 		if (IS_ERR(rt))
507 			goto err_free_skb;
508 		if (use_cache)
509 			dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst,
510 					  fl->saddr);
511 	}
512 
513 	min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
514 			+ tunnel_hlen + sizeof(struct iphdr);
515 	if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
516 		int head_delta = SKB_DATA_ALIGN(min_headroom -
517 						skb_headroom(skb) +
518 						16);
519 		err = pskb_expand_head(skb, max_t(int, head_delta, 0),
520 				       0, GFP_ATOMIC);
521 		if (unlikely(err))
522 			goto err_free_rt;
523 	}
524 	return rt;
525 
526 err_free_rt:
527 	ip_rt_put(rt);
528 err_free_skb:
529 	kfree_skb(skb);
530 	dev->stats.tx_dropped++;
531 	return NULL;
532 }
533 
534 static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev,
535 			__be16 proto)
536 {
537 	struct ip_tunnel_info *tun_info;
538 	const struct ip_tunnel_key *key;
539 	struct rtable *rt = NULL;
540 	struct flowi4 fl;
541 	int tunnel_hlen;
542 	__be16 df, flags;
543 
544 	tun_info = skb_tunnel_info(skb);
545 	if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
546 		     ip_tunnel_info_af(tun_info) != AF_INET))
547 		goto err_free_skb;
548 
549 	key = &tun_info->key;
550 	tunnel_hlen = gre_calc_hlen(key->tun_flags);
551 
552 	rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
553 	if (!rt)
554 		return;
555 
556 	/* Push Tunnel header. */
557 	if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM)))
558 		goto err_free_rt;
559 
560 	flags = tun_info->key.tun_flags & (TUNNEL_CSUM | TUNNEL_KEY);
561 	gre_build_header(skb, tunnel_hlen, flags, proto,
562 			 tunnel_id_to_key32(tun_info->key.tun_id), 0);
563 
564 	df = key->tun_flags & TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
565 
566 	iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
567 		      key->tos, key->ttl, df, false);
568 	return;
569 
570 err_free_rt:
571 	ip_rt_put(rt);
572 err_free_skb:
573 	kfree_skb(skb);
574 	dev->stats.tx_dropped++;
575 }
576 
577 static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev,
578 			   __be16 proto)
579 {
580 	struct ip_tunnel *tunnel = netdev_priv(dev);
581 	struct ip_tunnel_info *tun_info;
582 	const struct ip_tunnel_key *key;
583 	struct erspan_metadata *md;
584 	struct rtable *rt = NULL;
585 	bool truncate = false;
586 	struct flowi4 fl;
587 	int tunnel_hlen;
588 	int version;
589 	__be16 df;
590 
591 	tun_info = skb_tunnel_info(skb);
592 	if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
593 		     ip_tunnel_info_af(tun_info) != AF_INET))
594 		goto err_free_skb;
595 
596 	key = &tun_info->key;
597 	md = ip_tunnel_info_opts(tun_info);
598 	if (!md)
599 		goto err_free_rt;
600 
601 	/* ERSPAN has fixed 8 byte GRE header */
602 	version = md->version;
603 	tunnel_hlen = 8 + erspan_hdr_len(version);
604 
605 	rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
606 	if (!rt)
607 		return;
608 
609 	if (gre_handle_offloads(skb, false))
610 		goto err_free_rt;
611 
612 	if (skb->len > dev->mtu + dev->hard_header_len) {
613 		pskb_trim(skb, dev->mtu + dev->hard_header_len);
614 		truncate = true;
615 	}
616 
617 	if (version == 1) {
618 		erspan_build_header(skb, tunnel_id_to_key32(key->tun_id),
619 				    ntohl(md->u.index), truncate, true);
620 	} else if (version == 2) {
621 		u16 md2_flags;
622 		u8 direction;
623 		u16 hwid;
624 
625 		md2_flags = ntohs(md->u.md2.flags);
626 		direction = (md2_flags & DIR_MASK) >> DIR_OFFSET;
627 		hwid = (md2_flags & HWID_MASK) >> HWID_OFFSET;
628 
629 		erspan_build_header_v2(skb, tunnel_id_to_key32(key->tun_id),
630 				       direction, hwid,	truncate, true);
631 	} else {
632 		goto err_free_rt;
633 	}
634 
635 	gre_build_header(skb, 8, TUNNEL_SEQ,
636 			 htons(ETH_P_ERSPAN), 0, htonl(tunnel->o_seqno++));
637 
638 	df = key->tun_flags & TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
639 
640 	iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
641 		      key->tos, key->ttl, df, false);
642 	return;
643 
644 err_free_rt:
645 	ip_rt_put(rt);
646 err_free_skb:
647 	kfree_skb(skb);
648 	dev->stats.tx_dropped++;
649 }
650 
651 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
652 {
653 	struct ip_tunnel_info *info = skb_tunnel_info(skb);
654 	struct rtable *rt;
655 	struct flowi4 fl4;
656 
657 	if (ip_tunnel_info_af(info) != AF_INET)
658 		return -EINVAL;
659 
660 	rt = gre_get_rt(skb, dev, &fl4, &info->key);
661 	if (IS_ERR(rt))
662 		return PTR_ERR(rt);
663 
664 	ip_rt_put(rt);
665 	info->key.u.ipv4.src = fl4.saddr;
666 	return 0;
667 }
668 
669 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
670 			      struct net_device *dev)
671 {
672 	struct ip_tunnel *tunnel = netdev_priv(dev);
673 	const struct iphdr *tnl_params;
674 
675 	if (tunnel->collect_md) {
676 		gre_fb_xmit(skb, dev, skb->protocol);
677 		return NETDEV_TX_OK;
678 	}
679 
680 	if (dev->header_ops) {
681 		/* Need space for new headers */
682 		if (skb_cow_head(skb, dev->needed_headroom -
683 				      (tunnel->hlen + sizeof(struct iphdr))))
684 			goto free_skb;
685 
686 		tnl_params = (const struct iphdr *)skb->data;
687 
688 		/* Pull skb since ip_tunnel_xmit() needs skb->data pointing
689 		 * to gre header.
690 		 */
691 		skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
692 		skb_reset_mac_header(skb);
693 	} else {
694 		if (skb_cow_head(skb, dev->needed_headroom))
695 			goto free_skb;
696 
697 		tnl_params = &tunnel->parms.iph;
698 	}
699 
700 	if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
701 		goto free_skb;
702 
703 	__gre_xmit(skb, dev, tnl_params, skb->protocol);
704 	return NETDEV_TX_OK;
705 
706 free_skb:
707 	kfree_skb(skb);
708 	dev->stats.tx_dropped++;
709 	return NETDEV_TX_OK;
710 }
711 
712 static netdev_tx_t erspan_xmit(struct sk_buff *skb,
713 			       struct net_device *dev)
714 {
715 	struct ip_tunnel *tunnel = netdev_priv(dev);
716 	bool truncate = false;
717 
718 	if (tunnel->collect_md) {
719 		erspan_fb_xmit(skb, dev, skb->protocol);
720 		return NETDEV_TX_OK;
721 	}
722 
723 	if (gre_handle_offloads(skb, false))
724 		goto free_skb;
725 
726 	if (skb_cow_head(skb, dev->needed_headroom))
727 		goto free_skb;
728 
729 	if (skb->len > dev->mtu + dev->hard_header_len) {
730 		pskb_trim(skb, dev->mtu + dev->hard_header_len);
731 		truncate = true;
732 	}
733 
734 	/* Push ERSPAN header */
735 	if (tunnel->erspan_ver == 1)
736 		erspan_build_header(skb, tunnel->parms.o_key, tunnel->index,
737 				    truncate, true);
738 	else
739 		erspan_build_header_v2(skb, tunnel->parms.o_key,
740 				       tunnel->dir, tunnel->hwid,
741 				       truncate, true);
742 
743 	tunnel->parms.o_flags &= ~TUNNEL_KEY;
744 	__gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_ERSPAN));
745 	return NETDEV_TX_OK;
746 
747 free_skb:
748 	kfree_skb(skb);
749 	dev->stats.tx_dropped++;
750 	return NETDEV_TX_OK;
751 }
752 
753 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
754 				struct net_device *dev)
755 {
756 	struct ip_tunnel *tunnel = netdev_priv(dev);
757 
758 	if (tunnel->collect_md) {
759 		gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
760 		return NETDEV_TX_OK;
761 	}
762 
763 	if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
764 		goto free_skb;
765 
766 	if (skb_cow_head(skb, dev->needed_headroom))
767 		goto free_skb;
768 
769 	__gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
770 	return NETDEV_TX_OK;
771 
772 free_skb:
773 	kfree_skb(skb);
774 	dev->stats.tx_dropped++;
775 	return NETDEV_TX_OK;
776 }
777 
778 static void ipgre_link_update(struct net_device *dev, bool set_mtu)
779 {
780 	struct ip_tunnel *tunnel = netdev_priv(dev);
781 	int len;
782 
783 	len = tunnel->tun_hlen;
784 	tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
785 	len = tunnel->tun_hlen - len;
786 	tunnel->hlen = tunnel->hlen + len;
787 
788 	dev->needed_headroom = dev->needed_headroom + len;
789 	if (set_mtu)
790 		dev->mtu = max_t(int, dev->mtu - len, 68);
791 
792 	if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
793 		if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
794 		    tunnel->encap.type == TUNNEL_ENCAP_NONE) {
795 			dev->features |= NETIF_F_GSO_SOFTWARE;
796 			dev->hw_features |= NETIF_F_GSO_SOFTWARE;
797 		}
798 		dev->features |= NETIF_F_LLTX;
799 	}
800 }
801 
802 static int ipgre_tunnel_ioctl(struct net_device *dev,
803 			      struct ifreq *ifr, int cmd)
804 {
805 	struct ip_tunnel_parm p;
806 	int err;
807 
808 	if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
809 		return -EFAULT;
810 
811 	if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
812 		if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
813 		    p.iph.ihl != 5 || (p.iph.frag_off & htons(~IP_DF)) ||
814 		    ((p.i_flags | p.o_flags) & (GRE_VERSION | GRE_ROUTING)))
815 			return -EINVAL;
816 	}
817 
818 	p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
819 	p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
820 
821 	err = ip_tunnel_ioctl(dev, &p, cmd);
822 	if (err)
823 		return err;
824 
825 	if (cmd == SIOCCHGTUNNEL) {
826 		struct ip_tunnel *t = netdev_priv(dev);
827 
828 		t->parms.i_flags = p.i_flags;
829 		t->parms.o_flags = p.o_flags;
830 
831 		if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
832 			ipgre_link_update(dev, true);
833 	}
834 
835 	p.i_flags = gre_tnl_flags_to_gre_flags(p.i_flags);
836 	p.o_flags = gre_tnl_flags_to_gre_flags(p.o_flags);
837 
838 	if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
839 		return -EFAULT;
840 
841 	return 0;
842 }
843 
844 /* Nice toy. Unfortunately, useless in real life :-)
845    It allows to construct virtual multiprotocol broadcast "LAN"
846    over the Internet, provided multicast routing is tuned.
847 
848 
849    I have no idea was this bicycle invented before me,
850    so that I had to set ARPHRD_IPGRE to a random value.
851    I have an impression, that Cisco could make something similar,
852    but this feature is apparently missing in IOS<=11.2(8).
853 
854    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
855    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
856 
857    ping -t 255 224.66.66.66
858 
859    If nobody answers, mbone does not work.
860 
861    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
862    ip addr add 10.66.66.<somewhat>/24 dev Universe
863    ifconfig Universe up
864    ifconfig Universe add fe80::<Your_real_addr>/10
865    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
866    ftp 10.66.66.66
867    ...
868    ftp fec0:6666:6666::193.233.7.65
869    ...
870  */
871 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
872 			unsigned short type,
873 			const void *daddr, const void *saddr, unsigned int len)
874 {
875 	struct ip_tunnel *t = netdev_priv(dev);
876 	struct iphdr *iph;
877 	struct gre_base_hdr *greh;
878 
879 	iph = skb_push(skb, t->hlen + sizeof(*iph));
880 	greh = (struct gre_base_hdr *)(iph+1);
881 	greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
882 	greh->protocol = htons(type);
883 
884 	memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
885 
886 	/* Set the source hardware address. */
887 	if (saddr)
888 		memcpy(&iph->saddr, saddr, 4);
889 	if (daddr)
890 		memcpy(&iph->daddr, daddr, 4);
891 	if (iph->daddr)
892 		return t->hlen + sizeof(*iph);
893 
894 	return -(t->hlen + sizeof(*iph));
895 }
896 
897 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
898 {
899 	const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
900 	memcpy(haddr, &iph->saddr, 4);
901 	return 4;
902 }
903 
904 static const struct header_ops ipgre_header_ops = {
905 	.create	= ipgre_header,
906 	.parse	= ipgre_header_parse,
907 };
908 
909 #ifdef CONFIG_NET_IPGRE_BROADCAST
910 static int ipgre_open(struct net_device *dev)
911 {
912 	struct ip_tunnel *t = netdev_priv(dev);
913 
914 	if (ipv4_is_multicast(t->parms.iph.daddr)) {
915 		struct flowi4 fl4;
916 		struct rtable *rt;
917 
918 		rt = ip_route_output_gre(t->net, &fl4,
919 					 t->parms.iph.daddr,
920 					 t->parms.iph.saddr,
921 					 t->parms.o_key,
922 					 RT_TOS(t->parms.iph.tos),
923 					 t->parms.link);
924 		if (IS_ERR(rt))
925 			return -EADDRNOTAVAIL;
926 		dev = rt->dst.dev;
927 		ip_rt_put(rt);
928 		if (!__in_dev_get_rtnl(dev))
929 			return -EADDRNOTAVAIL;
930 		t->mlink = dev->ifindex;
931 		ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
932 	}
933 	return 0;
934 }
935 
936 static int ipgre_close(struct net_device *dev)
937 {
938 	struct ip_tunnel *t = netdev_priv(dev);
939 
940 	if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
941 		struct in_device *in_dev;
942 		in_dev = inetdev_by_index(t->net, t->mlink);
943 		if (in_dev)
944 			ip_mc_dec_group(in_dev, t->parms.iph.daddr);
945 	}
946 	return 0;
947 }
948 #endif
949 
950 static const struct net_device_ops ipgre_netdev_ops = {
951 	.ndo_init		= ipgre_tunnel_init,
952 	.ndo_uninit		= ip_tunnel_uninit,
953 #ifdef CONFIG_NET_IPGRE_BROADCAST
954 	.ndo_open		= ipgre_open,
955 	.ndo_stop		= ipgre_close,
956 #endif
957 	.ndo_start_xmit		= ipgre_xmit,
958 	.ndo_do_ioctl		= ipgre_tunnel_ioctl,
959 	.ndo_change_mtu		= ip_tunnel_change_mtu,
960 	.ndo_get_stats64	= ip_tunnel_get_stats64,
961 	.ndo_get_iflink		= ip_tunnel_get_iflink,
962 };
963 
964 #define GRE_FEATURES (NETIF_F_SG |		\
965 		      NETIF_F_FRAGLIST |	\
966 		      NETIF_F_HIGHDMA |		\
967 		      NETIF_F_HW_CSUM)
968 
969 static void ipgre_tunnel_setup(struct net_device *dev)
970 {
971 	dev->netdev_ops		= &ipgre_netdev_ops;
972 	dev->type		= ARPHRD_IPGRE;
973 	ip_tunnel_setup(dev, ipgre_net_id);
974 }
975 
976 static void __gre_tunnel_init(struct net_device *dev)
977 {
978 	struct ip_tunnel *tunnel;
979 	int t_hlen;
980 
981 	tunnel = netdev_priv(dev);
982 	tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
983 	tunnel->parms.iph.protocol = IPPROTO_GRE;
984 
985 	tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
986 
987 	t_hlen = tunnel->hlen + sizeof(struct iphdr);
988 
989 	dev->needed_headroom	= LL_MAX_HEADER + t_hlen + 4;
990 	dev->mtu		= ETH_DATA_LEN - t_hlen - 4;
991 
992 	dev->features		|= GRE_FEATURES;
993 	dev->hw_features	|= GRE_FEATURES;
994 
995 	if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
996 		/* TCP offload with GRE SEQ is not supported, nor
997 		 * can we support 2 levels of outer headers requiring
998 		 * an update.
999 		 */
1000 		if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
1001 		    (tunnel->encap.type == TUNNEL_ENCAP_NONE)) {
1002 			dev->features    |= NETIF_F_GSO_SOFTWARE;
1003 			dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1004 		}
1005 
1006 		/* Can use a lockless transmit, unless we generate
1007 		 * output sequences
1008 		 */
1009 		dev->features |= NETIF_F_LLTX;
1010 	}
1011 }
1012 
1013 static int ipgre_tunnel_init(struct net_device *dev)
1014 {
1015 	struct ip_tunnel *tunnel = netdev_priv(dev);
1016 	struct iphdr *iph = &tunnel->parms.iph;
1017 
1018 	__gre_tunnel_init(dev);
1019 
1020 	memcpy(dev->dev_addr, &iph->saddr, 4);
1021 	memcpy(dev->broadcast, &iph->daddr, 4);
1022 
1023 	dev->flags		= IFF_NOARP;
1024 	netif_keep_dst(dev);
1025 	dev->addr_len		= 4;
1026 
1027 	if (iph->daddr && !tunnel->collect_md) {
1028 #ifdef CONFIG_NET_IPGRE_BROADCAST
1029 		if (ipv4_is_multicast(iph->daddr)) {
1030 			if (!iph->saddr)
1031 				return -EINVAL;
1032 			dev->flags = IFF_BROADCAST;
1033 			dev->header_ops = &ipgre_header_ops;
1034 		}
1035 #endif
1036 	} else if (!tunnel->collect_md) {
1037 		dev->header_ops = &ipgre_header_ops;
1038 	}
1039 
1040 	return ip_tunnel_init(dev);
1041 }
1042 
1043 static const struct gre_protocol ipgre_protocol = {
1044 	.handler     = gre_rcv,
1045 	.err_handler = gre_err,
1046 };
1047 
1048 static int __net_init ipgre_init_net(struct net *net)
1049 {
1050 	return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
1051 }
1052 
1053 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net)
1054 {
1055 	ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops);
1056 }
1057 
1058 static struct pernet_operations ipgre_net_ops = {
1059 	.init = ipgre_init_net,
1060 	.exit_batch = ipgre_exit_batch_net,
1061 	.id   = &ipgre_net_id,
1062 	.size = sizeof(struct ip_tunnel_net),
1063 };
1064 
1065 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1066 				 struct netlink_ext_ack *extack)
1067 {
1068 	__be16 flags;
1069 
1070 	if (!data)
1071 		return 0;
1072 
1073 	flags = 0;
1074 	if (data[IFLA_GRE_IFLAGS])
1075 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1076 	if (data[IFLA_GRE_OFLAGS])
1077 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1078 	if (flags & (GRE_VERSION|GRE_ROUTING))
1079 		return -EINVAL;
1080 
1081 	if (data[IFLA_GRE_COLLECT_METADATA] &&
1082 	    data[IFLA_GRE_ENCAP_TYPE] &&
1083 	    nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1084 		return -EINVAL;
1085 
1086 	return 0;
1087 }
1088 
1089 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1090 			      struct netlink_ext_ack *extack)
1091 {
1092 	__be32 daddr;
1093 
1094 	if (tb[IFLA_ADDRESS]) {
1095 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1096 			return -EINVAL;
1097 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1098 			return -EADDRNOTAVAIL;
1099 	}
1100 
1101 	if (!data)
1102 		goto out;
1103 
1104 	if (data[IFLA_GRE_REMOTE]) {
1105 		memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1106 		if (!daddr)
1107 			return -EINVAL;
1108 	}
1109 
1110 out:
1111 	return ipgre_tunnel_validate(tb, data, extack);
1112 }
1113 
1114 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1115 			   struct netlink_ext_ack *extack)
1116 {
1117 	__be16 flags = 0;
1118 	int ret;
1119 
1120 	if (!data)
1121 		return 0;
1122 
1123 	ret = ipgre_tap_validate(tb, data, extack);
1124 	if (ret)
1125 		return ret;
1126 
1127 	/* ERSPAN should only have GRE sequence and key flag */
1128 	if (data[IFLA_GRE_OFLAGS])
1129 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1130 	if (data[IFLA_GRE_IFLAGS])
1131 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1132 	if (!data[IFLA_GRE_COLLECT_METADATA] &&
1133 	    flags != (GRE_SEQ | GRE_KEY))
1134 		return -EINVAL;
1135 
1136 	/* ERSPAN Session ID only has 10-bit. Since we reuse
1137 	 * 32-bit key field as ID, check it's range.
1138 	 */
1139 	if (data[IFLA_GRE_IKEY] &&
1140 	    (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1141 		return -EINVAL;
1142 
1143 	if (data[IFLA_GRE_OKEY] &&
1144 	    (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1145 		return -EINVAL;
1146 
1147 	return 0;
1148 }
1149 
1150 static int ipgre_netlink_parms(struct net_device *dev,
1151 				struct nlattr *data[],
1152 				struct nlattr *tb[],
1153 				struct ip_tunnel_parm *parms,
1154 				__u32 *fwmark)
1155 {
1156 	struct ip_tunnel *t = netdev_priv(dev);
1157 
1158 	memset(parms, 0, sizeof(*parms));
1159 
1160 	parms->iph.protocol = IPPROTO_GRE;
1161 
1162 	if (!data)
1163 		return 0;
1164 
1165 	if (data[IFLA_GRE_LINK])
1166 		parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1167 
1168 	if (data[IFLA_GRE_IFLAGS])
1169 		parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
1170 
1171 	if (data[IFLA_GRE_OFLAGS])
1172 		parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
1173 
1174 	if (data[IFLA_GRE_IKEY])
1175 		parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1176 
1177 	if (data[IFLA_GRE_OKEY])
1178 		parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1179 
1180 	if (data[IFLA_GRE_LOCAL])
1181 		parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1182 
1183 	if (data[IFLA_GRE_REMOTE])
1184 		parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1185 
1186 	if (data[IFLA_GRE_TTL])
1187 		parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1188 
1189 	if (data[IFLA_GRE_TOS])
1190 		parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1191 
1192 	if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1193 		if (t->ignore_df)
1194 			return -EINVAL;
1195 		parms->iph.frag_off = htons(IP_DF);
1196 	}
1197 
1198 	if (data[IFLA_GRE_COLLECT_METADATA]) {
1199 		t->collect_md = true;
1200 		if (dev->type == ARPHRD_IPGRE)
1201 			dev->type = ARPHRD_NONE;
1202 	}
1203 
1204 	if (data[IFLA_GRE_IGNORE_DF]) {
1205 		if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1206 		  && (parms->iph.frag_off & htons(IP_DF)))
1207 			return -EINVAL;
1208 		t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1209 	}
1210 
1211 	if (data[IFLA_GRE_FWMARK])
1212 		*fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1213 
1214 	if (data[IFLA_GRE_ERSPAN_VER]) {
1215 		t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]);
1216 
1217 		if (t->erspan_ver != 1 && t->erspan_ver != 2)
1218 			return -EINVAL;
1219 	}
1220 
1221 	if (t->erspan_ver == 1) {
1222 		if (data[IFLA_GRE_ERSPAN_INDEX]) {
1223 			t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1224 			if (t->index & ~INDEX_MASK)
1225 				return -EINVAL;
1226 		}
1227 	} else if (t->erspan_ver == 2) {
1228 		if (data[IFLA_GRE_ERSPAN_DIR]) {
1229 			t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]);
1230 			if (t->dir & ~(DIR_MASK >> DIR_OFFSET))
1231 				return -EINVAL;
1232 		}
1233 		if (data[IFLA_GRE_ERSPAN_HWID]) {
1234 			t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]);
1235 			if (t->hwid & ~(HWID_MASK >> HWID_OFFSET))
1236 				return -EINVAL;
1237 		}
1238 	}
1239 
1240 	return 0;
1241 }
1242 
1243 /* This function returns true when ENCAP attributes are present in the nl msg */
1244 static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1245 				      struct ip_tunnel_encap *ipencap)
1246 {
1247 	bool ret = false;
1248 
1249 	memset(ipencap, 0, sizeof(*ipencap));
1250 
1251 	if (!data)
1252 		return ret;
1253 
1254 	if (data[IFLA_GRE_ENCAP_TYPE]) {
1255 		ret = true;
1256 		ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1257 	}
1258 
1259 	if (data[IFLA_GRE_ENCAP_FLAGS]) {
1260 		ret = true;
1261 		ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1262 	}
1263 
1264 	if (data[IFLA_GRE_ENCAP_SPORT]) {
1265 		ret = true;
1266 		ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1267 	}
1268 
1269 	if (data[IFLA_GRE_ENCAP_DPORT]) {
1270 		ret = true;
1271 		ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1272 	}
1273 
1274 	return ret;
1275 }
1276 
1277 static int gre_tap_init(struct net_device *dev)
1278 {
1279 	__gre_tunnel_init(dev);
1280 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1281 	netif_keep_dst(dev);
1282 
1283 	return ip_tunnel_init(dev);
1284 }
1285 
1286 static const struct net_device_ops gre_tap_netdev_ops = {
1287 	.ndo_init		= gre_tap_init,
1288 	.ndo_uninit		= ip_tunnel_uninit,
1289 	.ndo_start_xmit		= gre_tap_xmit,
1290 	.ndo_set_mac_address 	= eth_mac_addr,
1291 	.ndo_validate_addr	= eth_validate_addr,
1292 	.ndo_change_mtu		= ip_tunnel_change_mtu,
1293 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1294 	.ndo_get_iflink		= ip_tunnel_get_iflink,
1295 	.ndo_fill_metadata_dst	= gre_fill_metadata_dst,
1296 };
1297 
1298 static int erspan_tunnel_init(struct net_device *dev)
1299 {
1300 	struct ip_tunnel *tunnel = netdev_priv(dev);
1301 	int t_hlen;
1302 
1303 	tunnel->tun_hlen = 8;
1304 	tunnel->parms.iph.protocol = IPPROTO_GRE;
1305 	tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1306 		       erspan_hdr_len(tunnel->erspan_ver);
1307 	t_hlen = tunnel->hlen + sizeof(struct iphdr);
1308 
1309 	dev->needed_headroom = LL_MAX_HEADER + t_hlen + 4;
1310 	dev->mtu = ETH_DATA_LEN - t_hlen - 4;
1311 	dev->features		|= GRE_FEATURES;
1312 	dev->hw_features	|= GRE_FEATURES;
1313 	dev->priv_flags		|= IFF_LIVE_ADDR_CHANGE;
1314 	netif_keep_dst(dev);
1315 
1316 	return ip_tunnel_init(dev);
1317 }
1318 
1319 static const struct net_device_ops erspan_netdev_ops = {
1320 	.ndo_init		= erspan_tunnel_init,
1321 	.ndo_uninit		= ip_tunnel_uninit,
1322 	.ndo_start_xmit		= erspan_xmit,
1323 	.ndo_set_mac_address	= eth_mac_addr,
1324 	.ndo_validate_addr	= eth_validate_addr,
1325 	.ndo_change_mtu		= ip_tunnel_change_mtu,
1326 	.ndo_get_stats64	= ip_tunnel_get_stats64,
1327 	.ndo_get_iflink		= ip_tunnel_get_iflink,
1328 	.ndo_fill_metadata_dst	= gre_fill_metadata_dst,
1329 };
1330 
1331 static void ipgre_tap_setup(struct net_device *dev)
1332 {
1333 	ether_setup(dev);
1334 	dev->max_mtu = 0;
1335 	dev->netdev_ops	= &gre_tap_netdev_ops;
1336 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1337 	dev->priv_flags	|= IFF_LIVE_ADDR_CHANGE;
1338 	ip_tunnel_setup(dev, gre_tap_net_id);
1339 }
1340 
1341 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1342 			 struct nlattr *tb[], struct nlattr *data[],
1343 			 struct netlink_ext_ack *extack)
1344 {
1345 	struct ip_tunnel_parm p;
1346 	struct ip_tunnel_encap ipencap;
1347 	__u32 fwmark = 0;
1348 	int err;
1349 
1350 	if (ipgre_netlink_encap_parms(data, &ipencap)) {
1351 		struct ip_tunnel *t = netdev_priv(dev);
1352 		err = ip_tunnel_encap_setup(t, &ipencap);
1353 
1354 		if (err < 0)
1355 			return err;
1356 	}
1357 
1358 	err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1359 	if (err < 0)
1360 		return err;
1361 	return ip_tunnel_newlink(dev, tb, &p, fwmark);
1362 }
1363 
1364 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1365 			    struct nlattr *data[],
1366 			    struct netlink_ext_ack *extack)
1367 {
1368 	struct ip_tunnel *t = netdev_priv(dev);
1369 	struct ip_tunnel_encap ipencap;
1370 	__u32 fwmark = t->fwmark;
1371 	struct ip_tunnel_parm p;
1372 	int err;
1373 
1374 	if (ipgre_netlink_encap_parms(data, &ipencap)) {
1375 		err = ip_tunnel_encap_setup(t, &ipencap);
1376 
1377 		if (err < 0)
1378 			return err;
1379 	}
1380 
1381 	err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1382 	if (err < 0)
1383 		return err;
1384 
1385 	err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1386 	if (err < 0)
1387 		return err;
1388 
1389 	t->parms.i_flags = p.i_flags;
1390 	t->parms.o_flags = p.o_flags;
1391 
1392 	if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
1393 		ipgre_link_update(dev, !tb[IFLA_MTU]);
1394 
1395 	return 0;
1396 }
1397 
1398 static size_t ipgre_get_size(const struct net_device *dev)
1399 {
1400 	return
1401 		/* IFLA_GRE_LINK */
1402 		nla_total_size(4) +
1403 		/* IFLA_GRE_IFLAGS */
1404 		nla_total_size(2) +
1405 		/* IFLA_GRE_OFLAGS */
1406 		nla_total_size(2) +
1407 		/* IFLA_GRE_IKEY */
1408 		nla_total_size(4) +
1409 		/* IFLA_GRE_OKEY */
1410 		nla_total_size(4) +
1411 		/* IFLA_GRE_LOCAL */
1412 		nla_total_size(4) +
1413 		/* IFLA_GRE_REMOTE */
1414 		nla_total_size(4) +
1415 		/* IFLA_GRE_TTL */
1416 		nla_total_size(1) +
1417 		/* IFLA_GRE_TOS */
1418 		nla_total_size(1) +
1419 		/* IFLA_GRE_PMTUDISC */
1420 		nla_total_size(1) +
1421 		/* IFLA_GRE_ENCAP_TYPE */
1422 		nla_total_size(2) +
1423 		/* IFLA_GRE_ENCAP_FLAGS */
1424 		nla_total_size(2) +
1425 		/* IFLA_GRE_ENCAP_SPORT */
1426 		nla_total_size(2) +
1427 		/* IFLA_GRE_ENCAP_DPORT */
1428 		nla_total_size(2) +
1429 		/* IFLA_GRE_COLLECT_METADATA */
1430 		nla_total_size(0) +
1431 		/* IFLA_GRE_IGNORE_DF */
1432 		nla_total_size(1) +
1433 		/* IFLA_GRE_FWMARK */
1434 		nla_total_size(4) +
1435 		/* IFLA_GRE_ERSPAN_INDEX */
1436 		nla_total_size(4) +
1437 		/* IFLA_GRE_ERSPAN_VER */
1438 		nla_total_size(1) +
1439 		/* IFLA_GRE_ERSPAN_DIR */
1440 		nla_total_size(1) +
1441 		/* IFLA_GRE_ERSPAN_HWID */
1442 		nla_total_size(2) +
1443 		0;
1444 }
1445 
1446 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1447 {
1448 	struct ip_tunnel *t = netdev_priv(dev);
1449 	struct ip_tunnel_parm *p = &t->parms;
1450 
1451 	if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1452 	    nla_put_be16(skb, IFLA_GRE_IFLAGS,
1453 			 gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1454 	    nla_put_be16(skb, IFLA_GRE_OFLAGS,
1455 			 gre_tnl_flags_to_gre_flags(p->o_flags)) ||
1456 	    nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1457 	    nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1458 	    nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1459 	    nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1460 	    nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1461 	    nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1462 	    nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1463 		       !!(p->iph.frag_off & htons(IP_DF))) ||
1464 	    nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1465 		goto nla_put_failure;
1466 
1467 	if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1468 			t->encap.type) ||
1469 	    nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1470 			 t->encap.sport) ||
1471 	    nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1472 			 t->encap.dport) ||
1473 	    nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1474 			t->encap.flags))
1475 		goto nla_put_failure;
1476 
1477 	if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1478 		goto nla_put_failure;
1479 
1480 	if (t->collect_md) {
1481 		if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1482 			goto nla_put_failure;
1483 	}
1484 
1485 	if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1486 		goto nla_put_failure;
1487 
1488 	if (t->erspan_ver == 1) {
1489 		if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1490 			goto nla_put_failure;
1491 	} else if (t->erspan_ver == 2) {
1492 		if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1493 			goto nla_put_failure;
1494 		if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1495 			goto nla_put_failure;
1496 	}
1497 
1498 	return 0;
1499 
1500 nla_put_failure:
1501 	return -EMSGSIZE;
1502 }
1503 
1504 static void erspan_setup(struct net_device *dev)
1505 {
1506 	ether_setup(dev);
1507 	dev->netdev_ops = &erspan_netdev_ops;
1508 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1509 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1510 	ip_tunnel_setup(dev, erspan_net_id);
1511 }
1512 
1513 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1514 	[IFLA_GRE_LINK]		= { .type = NLA_U32 },
1515 	[IFLA_GRE_IFLAGS]	= { .type = NLA_U16 },
1516 	[IFLA_GRE_OFLAGS]	= { .type = NLA_U16 },
1517 	[IFLA_GRE_IKEY]		= { .type = NLA_U32 },
1518 	[IFLA_GRE_OKEY]		= { .type = NLA_U32 },
1519 	[IFLA_GRE_LOCAL]	= { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1520 	[IFLA_GRE_REMOTE]	= { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1521 	[IFLA_GRE_TTL]		= { .type = NLA_U8 },
1522 	[IFLA_GRE_TOS]		= { .type = NLA_U8 },
1523 	[IFLA_GRE_PMTUDISC]	= { .type = NLA_U8 },
1524 	[IFLA_GRE_ENCAP_TYPE]	= { .type = NLA_U16 },
1525 	[IFLA_GRE_ENCAP_FLAGS]	= { .type = NLA_U16 },
1526 	[IFLA_GRE_ENCAP_SPORT]	= { .type = NLA_U16 },
1527 	[IFLA_GRE_ENCAP_DPORT]	= { .type = NLA_U16 },
1528 	[IFLA_GRE_COLLECT_METADATA]	= { .type = NLA_FLAG },
1529 	[IFLA_GRE_IGNORE_DF]	= { .type = NLA_U8 },
1530 	[IFLA_GRE_FWMARK]	= { .type = NLA_U32 },
1531 	[IFLA_GRE_ERSPAN_INDEX]	= { .type = NLA_U32 },
1532 	[IFLA_GRE_ERSPAN_VER]	= { .type = NLA_U8 },
1533 	[IFLA_GRE_ERSPAN_DIR]	= { .type = NLA_U8 },
1534 	[IFLA_GRE_ERSPAN_HWID]	= { .type = NLA_U16 },
1535 };
1536 
1537 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1538 	.kind		= "gre",
1539 	.maxtype	= IFLA_GRE_MAX,
1540 	.policy		= ipgre_policy,
1541 	.priv_size	= sizeof(struct ip_tunnel),
1542 	.setup		= ipgre_tunnel_setup,
1543 	.validate	= ipgre_tunnel_validate,
1544 	.newlink	= ipgre_newlink,
1545 	.changelink	= ipgre_changelink,
1546 	.dellink	= ip_tunnel_dellink,
1547 	.get_size	= ipgre_get_size,
1548 	.fill_info	= ipgre_fill_info,
1549 	.get_link_net	= ip_tunnel_get_link_net,
1550 };
1551 
1552 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1553 	.kind		= "gretap",
1554 	.maxtype	= IFLA_GRE_MAX,
1555 	.policy		= ipgre_policy,
1556 	.priv_size	= sizeof(struct ip_tunnel),
1557 	.setup		= ipgre_tap_setup,
1558 	.validate	= ipgre_tap_validate,
1559 	.newlink	= ipgre_newlink,
1560 	.changelink	= ipgre_changelink,
1561 	.dellink	= ip_tunnel_dellink,
1562 	.get_size	= ipgre_get_size,
1563 	.fill_info	= ipgre_fill_info,
1564 	.get_link_net	= ip_tunnel_get_link_net,
1565 };
1566 
1567 static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1568 	.kind		= "erspan",
1569 	.maxtype	= IFLA_GRE_MAX,
1570 	.policy		= ipgre_policy,
1571 	.priv_size	= sizeof(struct ip_tunnel),
1572 	.setup		= erspan_setup,
1573 	.validate	= erspan_validate,
1574 	.newlink	= ipgre_newlink,
1575 	.changelink	= ipgre_changelink,
1576 	.dellink	= ip_tunnel_dellink,
1577 	.get_size	= ipgre_get_size,
1578 	.fill_info	= ipgre_fill_info,
1579 	.get_link_net	= ip_tunnel_get_link_net,
1580 };
1581 
1582 struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1583 					u8 name_assign_type)
1584 {
1585 	struct nlattr *tb[IFLA_MAX + 1];
1586 	struct net_device *dev;
1587 	LIST_HEAD(list_kill);
1588 	struct ip_tunnel *t;
1589 	int err;
1590 
1591 	memset(&tb, 0, sizeof(tb));
1592 
1593 	dev = rtnl_create_link(net, name, name_assign_type,
1594 			       &ipgre_tap_ops, tb);
1595 	if (IS_ERR(dev))
1596 		return dev;
1597 
1598 	/* Configure flow based GRE device. */
1599 	t = netdev_priv(dev);
1600 	t->collect_md = true;
1601 
1602 	err = ipgre_newlink(net, dev, tb, NULL, NULL);
1603 	if (err < 0) {
1604 		free_netdev(dev);
1605 		return ERR_PTR(err);
1606 	}
1607 
1608 	/* openvswitch users expect packet sizes to be unrestricted,
1609 	 * so set the largest MTU we can.
1610 	 */
1611 	err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1612 	if (err)
1613 		goto out;
1614 
1615 	err = rtnl_configure_link(dev, NULL);
1616 	if (err < 0)
1617 		goto out;
1618 
1619 	return dev;
1620 out:
1621 	ip_tunnel_dellink(dev, &list_kill);
1622 	unregister_netdevice_many(&list_kill);
1623 	return ERR_PTR(err);
1624 }
1625 EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1626 
1627 static int __net_init ipgre_tap_init_net(struct net *net)
1628 {
1629 	return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1630 }
1631 
1632 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net)
1633 {
1634 	ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops);
1635 }
1636 
1637 static struct pernet_operations ipgre_tap_net_ops = {
1638 	.init = ipgre_tap_init_net,
1639 	.exit_batch = ipgre_tap_exit_batch_net,
1640 	.id   = &gre_tap_net_id,
1641 	.size = sizeof(struct ip_tunnel_net),
1642 };
1643 
1644 static int __net_init erspan_init_net(struct net *net)
1645 {
1646 	return ip_tunnel_init_net(net, erspan_net_id,
1647 				  &erspan_link_ops, "erspan0");
1648 }
1649 
1650 static void __net_exit erspan_exit_batch_net(struct list_head *net_list)
1651 {
1652 	ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops);
1653 }
1654 
1655 static struct pernet_operations erspan_net_ops = {
1656 	.init = erspan_init_net,
1657 	.exit_batch = erspan_exit_batch_net,
1658 	.id   = &erspan_net_id,
1659 	.size = sizeof(struct ip_tunnel_net),
1660 };
1661 
1662 static int __init ipgre_init(void)
1663 {
1664 	int err;
1665 
1666 	pr_info("GRE over IPv4 tunneling driver\n");
1667 
1668 	err = register_pernet_device(&ipgre_net_ops);
1669 	if (err < 0)
1670 		return err;
1671 
1672 	err = register_pernet_device(&ipgre_tap_net_ops);
1673 	if (err < 0)
1674 		goto pnet_tap_failed;
1675 
1676 	err = register_pernet_device(&erspan_net_ops);
1677 	if (err < 0)
1678 		goto pnet_erspan_failed;
1679 
1680 	err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1681 	if (err < 0) {
1682 		pr_info("%s: can't add protocol\n", __func__);
1683 		goto add_proto_failed;
1684 	}
1685 
1686 	err = rtnl_link_register(&ipgre_link_ops);
1687 	if (err < 0)
1688 		goto rtnl_link_failed;
1689 
1690 	err = rtnl_link_register(&ipgre_tap_ops);
1691 	if (err < 0)
1692 		goto tap_ops_failed;
1693 
1694 	err = rtnl_link_register(&erspan_link_ops);
1695 	if (err < 0)
1696 		goto erspan_link_failed;
1697 
1698 	return 0;
1699 
1700 erspan_link_failed:
1701 	rtnl_link_unregister(&ipgre_tap_ops);
1702 tap_ops_failed:
1703 	rtnl_link_unregister(&ipgre_link_ops);
1704 rtnl_link_failed:
1705 	gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1706 add_proto_failed:
1707 	unregister_pernet_device(&erspan_net_ops);
1708 pnet_erspan_failed:
1709 	unregister_pernet_device(&ipgre_tap_net_ops);
1710 pnet_tap_failed:
1711 	unregister_pernet_device(&ipgre_net_ops);
1712 	return err;
1713 }
1714 
1715 static void __exit ipgre_fini(void)
1716 {
1717 	rtnl_link_unregister(&ipgre_tap_ops);
1718 	rtnl_link_unregister(&ipgre_link_ops);
1719 	rtnl_link_unregister(&erspan_link_ops);
1720 	gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1721 	unregister_pernet_device(&ipgre_tap_net_ops);
1722 	unregister_pernet_device(&ipgre_net_ops);
1723 	unregister_pernet_device(&erspan_net_ops);
1724 }
1725 
1726 module_init(ipgre_init);
1727 module_exit(ipgre_fini);
1728 MODULE_LICENSE("GPL");
1729 MODULE_ALIAS_RTNL_LINK("gre");
1730 MODULE_ALIAS_RTNL_LINK("gretap");
1731 MODULE_ALIAS_RTNL_LINK("erspan");
1732 MODULE_ALIAS_NETDEV("gre0");
1733 MODULE_ALIAS_NETDEV("gretap0");
1734 MODULE_ALIAS_NETDEV("erspan0");
1735