xref: /openbmc/linux/net/ipv4/ip_gre.c (revision 4161b450)
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 <asm/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/mroute.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 
51 #if IS_ENABLED(CONFIG_IPV6)
52 #include <net/ipv6.h>
53 #include <net/ip6_fib.h>
54 #include <net/ip6_route.h>
55 #endif
56 
57 /*
58    Problems & solutions
59    --------------------
60 
61    1. The most important issue is detecting local dead loops.
62    They would cause complete host lockup in transmit, which
63    would be "resolved" by stack overflow or, if queueing is enabled,
64    with infinite looping in net_bh.
65 
66    We cannot track such dead loops during route installation,
67    it is infeasible task. The most general solutions would be
68    to keep skb->encapsulation counter (sort of local ttl),
69    and silently drop packet when it expires. It is a good
70    solution, but it supposes maintaining new variable in ALL
71    skb, even if no tunneling is used.
72 
73    Current solution: xmit_recursion breaks dead loops. This is a percpu
74    counter, since when we enter the first ndo_xmit(), cpu migration is
75    forbidden. We force an exit if this counter reaches RECURSION_LIMIT
76 
77    2. Networking dead loops would not kill routers, but would really
78    kill network. IP hop limit plays role of "t->recursion" in this case,
79    if we copy it from packet being encapsulated to upper header.
80    It is very good solution, but it introduces two problems:
81 
82    - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
83      do not work over tunnels.
84    - traceroute does not work. I planned to relay ICMP from tunnel,
85      so that this problem would be solved and traceroute output
86      would even more informative. This idea appeared to be wrong:
87      only Linux complies to rfc1812 now (yes, guys, Linux is the only
88      true router now :-)), all routers (at least, in neighbourhood of mine)
89      return only 8 bytes of payload. It is the end.
90 
91    Hence, if we want that OSPF worked or traceroute said something reasonable,
92    we should search for another solution.
93 
94    One of them is to parse packet trying to detect inner encapsulation
95    made by our node. It is difficult or even impossible, especially,
96    taking into account fragmentation. TO be short, ttl is not solution at all.
97 
98    Current solution: The solution was UNEXPECTEDLY SIMPLE.
99    We force DF flag on tunnels with preconfigured hop limit,
100    that is ALL. :-) Well, it does not remove the problem completely,
101    but exponential growth of network traffic is changed to linear
102    (branches, that exceed pmtu are pruned) and tunnel mtu
103    rapidly degrades to value <68, where looping stops.
104    Yes, it is not good if there exists a router in the loop,
105    which does not force DF, even when encapsulating packets have DF set.
106    But it is not our problem! Nobody could accuse us, we made
107    all that we could make. Even if it is your gated who injected
108    fatal route to network, even if it were you who configured
109    fatal static route: you are innocent. :-)
110 
111    Alexey Kuznetsov.
112  */
113 
114 static bool log_ecn_error = true;
115 module_param(log_ecn_error, bool, 0644);
116 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
117 
118 static struct rtnl_link_ops ipgre_link_ops __read_mostly;
119 static int ipgre_tunnel_init(struct net_device *dev);
120 
121 static int ipgre_net_id __read_mostly;
122 static int gre_tap_net_id __read_mostly;
123 
124 static int 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 	struct ip_tunnel *t;
147 
148 	switch (type) {
149 	default:
150 	case ICMP_PARAMETERPROB:
151 		return PACKET_RCVD;
152 
153 	case ICMP_DEST_UNREACH:
154 		switch (code) {
155 		case ICMP_SR_FAILED:
156 		case ICMP_PORT_UNREACH:
157 			/* Impossible event. */
158 			return PACKET_RCVD;
159 		default:
160 			/* All others are translated to HOST_UNREACH.
161 			   rfc2003 contains "deep thoughts" about NET_UNREACH,
162 			   I believe they are just ether pollution. --ANK
163 			 */
164 			break;
165 		}
166 		break;
167 	case ICMP_TIME_EXCEEDED:
168 		if (code != ICMP_EXC_TTL)
169 			return PACKET_RCVD;
170 		break;
171 
172 	case ICMP_REDIRECT:
173 		break;
174 	}
175 
176 	if (tpi->proto == htons(ETH_P_TEB))
177 		itn = net_generic(net, gre_tap_net_id);
178 	else
179 		itn = net_generic(net, ipgre_net_id);
180 
181 	iph = (const struct iphdr *)(icmp_hdr(skb) + 1);
182 	t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
183 			     iph->daddr, iph->saddr, tpi->key);
184 
185 	if (t == NULL)
186 		return PACKET_REJECT;
187 
188 	if (t->parms.iph.daddr == 0 ||
189 	    ipv4_is_multicast(t->parms.iph.daddr))
190 		return PACKET_RCVD;
191 
192 	if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
193 		return PACKET_RCVD;
194 
195 	if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
196 		t->err_count++;
197 	else
198 		t->err_count = 1;
199 	t->err_time = jiffies;
200 	return PACKET_RCVD;
201 }
202 
203 static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi)
204 {
205 	struct net *net = dev_net(skb->dev);
206 	struct ip_tunnel_net *itn;
207 	const struct iphdr *iph;
208 	struct ip_tunnel *tunnel;
209 
210 	if (tpi->proto == htons(ETH_P_TEB))
211 		itn = net_generic(net, gre_tap_net_id);
212 	else
213 		itn = net_generic(net, ipgre_net_id);
214 
215 	iph = ip_hdr(skb);
216 	tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
217 				  iph->saddr, iph->daddr, tpi->key);
218 
219 	if (tunnel) {
220 		skb_pop_mac_header(skb);
221 		ip_tunnel_rcv(tunnel, skb, tpi, log_ecn_error);
222 		return PACKET_RCVD;
223 	}
224 	return PACKET_REJECT;
225 }
226 
227 static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
228 		       const struct iphdr *tnl_params,
229 		       __be16 proto)
230 {
231 	struct ip_tunnel *tunnel = netdev_priv(dev);
232 	struct tnl_ptk_info tpi;
233 
234 	tpi.flags = tunnel->parms.o_flags;
235 	tpi.proto = proto;
236 	tpi.key = tunnel->parms.o_key;
237 	if (tunnel->parms.o_flags & TUNNEL_SEQ)
238 		tunnel->o_seqno++;
239 	tpi.seq = htonl(tunnel->o_seqno);
240 
241 	/* Push GRE header. */
242 	gre_build_header(skb, &tpi, tunnel->tun_hlen);
243 
244 	skb_set_inner_protocol(skb, tpi.proto);
245 
246 	ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
247 }
248 
249 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
250 			      struct net_device *dev)
251 {
252 	struct ip_tunnel *tunnel = netdev_priv(dev);
253 	const struct iphdr *tnl_params;
254 
255 	if (dev->header_ops) {
256 		/* Need space for new headers */
257 		if (skb_cow_head(skb, dev->needed_headroom -
258 				      (tunnel->hlen + sizeof(struct iphdr))))
259 			goto free_skb;
260 
261 		tnl_params = (const struct iphdr *)skb->data;
262 
263 		/* Pull skb since ip_tunnel_xmit() needs skb->data pointing
264 		 * to gre header.
265 		 */
266 		skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
267 		skb_reset_mac_header(skb);
268 	} else {
269 		if (skb_cow_head(skb, dev->needed_headroom))
270 			goto free_skb;
271 
272 		tnl_params = &tunnel->parms.iph;
273 	}
274 
275 	skb = gre_handle_offloads(skb, !!(tunnel->parms.o_flags&TUNNEL_CSUM));
276 	if (IS_ERR(skb))
277 		goto out;
278 
279 	__gre_xmit(skb, dev, tnl_params, skb->protocol);
280 
281 	return NETDEV_TX_OK;
282 
283 free_skb:
284 	kfree_skb(skb);
285 out:
286 	dev->stats.tx_dropped++;
287 	return NETDEV_TX_OK;
288 }
289 
290 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
291 				struct net_device *dev)
292 {
293 	struct ip_tunnel *tunnel = netdev_priv(dev);
294 
295 	skb = gre_handle_offloads(skb, !!(tunnel->parms.o_flags&TUNNEL_CSUM));
296 	if (IS_ERR(skb))
297 		goto out;
298 
299 	if (skb_cow_head(skb, dev->needed_headroom))
300 		goto free_skb;
301 
302 	__gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
303 
304 	return NETDEV_TX_OK;
305 
306 free_skb:
307 	kfree_skb(skb);
308 out:
309 	dev->stats.tx_dropped++;
310 	return NETDEV_TX_OK;
311 }
312 
313 static int ipgre_tunnel_ioctl(struct net_device *dev,
314 			      struct ifreq *ifr, int cmd)
315 {
316 	int err;
317 	struct ip_tunnel_parm p;
318 
319 	if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
320 		return -EFAULT;
321 	if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
322 		if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
323 		    p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) ||
324 		    ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING)))
325 			return -EINVAL;
326 	}
327 	p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
328 	p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
329 
330 	err = ip_tunnel_ioctl(dev, &p, cmd);
331 	if (err)
332 		return err;
333 
334 	p.i_flags = tnl_flags_to_gre_flags(p.i_flags);
335 	p.o_flags = tnl_flags_to_gre_flags(p.o_flags);
336 
337 	if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
338 		return -EFAULT;
339 	return 0;
340 }
341 
342 /* Nice toy. Unfortunately, useless in real life :-)
343    It allows to construct virtual multiprotocol broadcast "LAN"
344    over the Internet, provided multicast routing is tuned.
345 
346 
347    I have no idea was this bicycle invented before me,
348    so that I had to set ARPHRD_IPGRE to a random value.
349    I have an impression, that Cisco could make something similar,
350    but this feature is apparently missing in IOS<=11.2(8).
351 
352    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
353    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
354 
355    ping -t 255 224.66.66.66
356 
357    If nobody answers, mbone does not work.
358 
359    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
360    ip addr add 10.66.66.<somewhat>/24 dev Universe
361    ifconfig Universe up
362    ifconfig Universe add fe80::<Your_real_addr>/10
363    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
364    ftp 10.66.66.66
365    ...
366    ftp fec0:6666:6666::193.233.7.65
367    ...
368  */
369 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
370 			unsigned short type,
371 			const void *daddr, const void *saddr, unsigned int len)
372 {
373 	struct ip_tunnel *t = netdev_priv(dev);
374 	struct iphdr *iph;
375 	struct gre_base_hdr *greh;
376 
377 	iph = (struct iphdr *)skb_push(skb, t->hlen + sizeof(*iph));
378 	greh = (struct gre_base_hdr *)(iph+1);
379 	greh->flags = tnl_flags_to_gre_flags(t->parms.o_flags);
380 	greh->protocol = htons(type);
381 
382 	memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
383 
384 	/* Set the source hardware address. */
385 	if (saddr)
386 		memcpy(&iph->saddr, saddr, 4);
387 	if (daddr)
388 		memcpy(&iph->daddr, daddr, 4);
389 	if (iph->daddr)
390 		return t->hlen + sizeof(*iph);
391 
392 	return -(t->hlen + sizeof(*iph));
393 }
394 
395 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
396 {
397 	const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
398 	memcpy(haddr, &iph->saddr, 4);
399 	return 4;
400 }
401 
402 static const struct header_ops ipgre_header_ops = {
403 	.create	= ipgre_header,
404 	.parse	= ipgre_header_parse,
405 };
406 
407 #ifdef CONFIG_NET_IPGRE_BROADCAST
408 static int ipgre_open(struct net_device *dev)
409 {
410 	struct ip_tunnel *t = netdev_priv(dev);
411 
412 	if (ipv4_is_multicast(t->parms.iph.daddr)) {
413 		struct flowi4 fl4;
414 		struct rtable *rt;
415 
416 		rt = ip_route_output_gre(t->net, &fl4,
417 					 t->parms.iph.daddr,
418 					 t->parms.iph.saddr,
419 					 t->parms.o_key,
420 					 RT_TOS(t->parms.iph.tos),
421 					 t->parms.link);
422 		if (IS_ERR(rt))
423 			return -EADDRNOTAVAIL;
424 		dev = rt->dst.dev;
425 		ip_rt_put(rt);
426 		if (__in_dev_get_rtnl(dev) == NULL)
427 			return -EADDRNOTAVAIL;
428 		t->mlink = dev->ifindex;
429 		ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
430 	}
431 	return 0;
432 }
433 
434 static int ipgre_close(struct net_device *dev)
435 {
436 	struct ip_tunnel *t = netdev_priv(dev);
437 
438 	if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
439 		struct in_device *in_dev;
440 		in_dev = inetdev_by_index(t->net, t->mlink);
441 		if (in_dev)
442 			ip_mc_dec_group(in_dev, t->parms.iph.daddr);
443 	}
444 	return 0;
445 }
446 #endif
447 
448 static const struct net_device_ops ipgre_netdev_ops = {
449 	.ndo_init		= ipgre_tunnel_init,
450 	.ndo_uninit		= ip_tunnel_uninit,
451 #ifdef CONFIG_NET_IPGRE_BROADCAST
452 	.ndo_open		= ipgre_open,
453 	.ndo_stop		= ipgre_close,
454 #endif
455 	.ndo_start_xmit		= ipgre_xmit,
456 	.ndo_do_ioctl		= ipgre_tunnel_ioctl,
457 	.ndo_change_mtu		= ip_tunnel_change_mtu,
458 	.ndo_get_stats64	= ip_tunnel_get_stats64,
459 };
460 
461 #define GRE_FEATURES (NETIF_F_SG |		\
462 		      NETIF_F_FRAGLIST |	\
463 		      NETIF_F_HIGHDMA |		\
464 		      NETIF_F_HW_CSUM)
465 
466 static void ipgre_tunnel_setup(struct net_device *dev)
467 {
468 	dev->netdev_ops		= &ipgre_netdev_ops;
469 	dev->type		= ARPHRD_IPGRE;
470 	ip_tunnel_setup(dev, ipgre_net_id);
471 }
472 
473 static void __gre_tunnel_init(struct net_device *dev)
474 {
475 	struct ip_tunnel *tunnel;
476 	int t_hlen;
477 
478 	tunnel = netdev_priv(dev);
479 	tunnel->tun_hlen = ip_gre_calc_hlen(tunnel->parms.o_flags);
480 	tunnel->parms.iph.protocol = IPPROTO_GRE;
481 
482 	tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
483 
484 	t_hlen = tunnel->hlen + sizeof(struct iphdr);
485 
486 	dev->needed_headroom	= LL_MAX_HEADER + t_hlen + 4;
487 	dev->mtu		= ETH_DATA_LEN - t_hlen - 4;
488 
489 	dev->features		|= GRE_FEATURES;
490 	dev->hw_features	|= GRE_FEATURES;
491 
492 	if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
493 		/* TCP offload with GRE SEQ is not supported. */
494 		dev->features    |= NETIF_F_GSO_SOFTWARE;
495 		dev->hw_features |= NETIF_F_GSO_SOFTWARE;
496 		/* Can use a lockless transmit, unless we generate
497 		 * output sequences
498 		 */
499 		dev->features |= NETIF_F_LLTX;
500 	}
501 }
502 
503 static int ipgre_tunnel_init(struct net_device *dev)
504 {
505 	struct ip_tunnel *tunnel = netdev_priv(dev);
506 	struct iphdr *iph = &tunnel->parms.iph;
507 
508 	__gre_tunnel_init(dev);
509 
510 	memcpy(dev->dev_addr, &iph->saddr, 4);
511 	memcpy(dev->broadcast, &iph->daddr, 4);
512 
513 	dev->flags		= IFF_NOARP;
514 	netif_keep_dst(dev);
515 	dev->addr_len		= 4;
516 
517 	if (iph->daddr) {
518 #ifdef CONFIG_NET_IPGRE_BROADCAST
519 		if (ipv4_is_multicast(iph->daddr)) {
520 			if (!iph->saddr)
521 				return -EINVAL;
522 			dev->flags = IFF_BROADCAST;
523 			dev->header_ops = &ipgre_header_ops;
524 		}
525 #endif
526 	} else
527 		dev->header_ops = &ipgre_header_ops;
528 
529 	return ip_tunnel_init(dev);
530 }
531 
532 static struct gre_cisco_protocol ipgre_protocol = {
533 	.handler        = ipgre_rcv,
534 	.err_handler    = ipgre_err,
535 	.priority       = 0,
536 };
537 
538 static int __net_init ipgre_init_net(struct net *net)
539 {
540 	return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
541 }
542 
543 static void __net_exit ipgre_exit_net(struct net *net)
544 {
545 	struct ip_tunnel_net *itn = net_generic(net, ipgre_net_id);
546 	ip_tunnel_delete_net(itn, &ipgre_link_ops);
547 }
548 
549 static struct pernet_operations ipgre_net_ops = {
550 	.init = ipgre_init_net,
551 	.exit = ipgre_exit_net,
552 	.id   = &ipgre_net_id,
553 	.size = sizeof(struct ip_tunnel_net),
554 };
555 
556 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
557 {
558 	__be16 flags;
559 
560 	if (!data)
561 		return 0;
562 
563 	flags = 0;
564 	if (data[IFLA_GRE_IFLAGS])
565 		flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
566 	if (data[IFLA_GRE_OFLAGS])
567 		flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
568 	if (flags & (GRE_VERSION|GRE_ROUTING))
569 		return -EINVAL;
570 
571 	return 0;
572 }
573 
574 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
575 {
576 	__be32 daddr;
577 
578 	if (tb[IFLA_ADDRESS]) {
579 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
580 			return -EINVAL;
581 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
582 			return -EADDRNOTAVAIL;
583 	}
584 
585 	if (!data)
586 		goto out;
587 
588 	if (data[IFLA_GRE_REMOTE]) {
589 		memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
590 		if (!daddr)
591 			return -EINVAL;
592 	}
593 
594 out:
595 	return ipgre_tunnel_validate(tb, data);
596 }
597 
598 static void ipgre_netlink_parms(struct nlattr *data[], struct nlattr *tb[],
599 			       struct ip_tunnel_parm *parms)
600 {
601 	memset(parms, 0, sizeof(*parms));
602 
603 	parms->iph.protocol = IPPROTO_GRE;
604 
605 	if (!data)
606 		return;
607 
608 	if (data[IFLA_GRE_LINK])
609 		parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
610 
611 	if (data[IFLA_GRE_IFLAGS])
612 		parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
613 
614 	if (data[IFLA_GRE_OFLAGS])
615 		parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
616 
617 	if (data[IFLA_GRE_IKEY])
618 		parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
619 
620 	if (data[IFLA_GRE_OKEY])
621 		parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
622 
623 	if (data[IFLA_GRE_LOCAL])
624 		parms->iph.saddr = nla_get_be32(data[IFLA_GRE_LOCAL]);
625 
626 	if (data[IFLA_GRE_REMOTE])
627 		parms->iph.daddr = nla_get_be32(data[IFLA_GRE_REMOTE]);
628 
629 	if (data[IFLA_GRE_TTL])
630 		parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
631 
632 	if (data[IFLA_GRE_TOS])
633 		parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
634 
635 	if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC]))
636 		parms->iph.frag_off = htons(IP_DF);
637 }
638 
639 /* This function returns true when ENCAP attributes are present in the nl msg */
640 static bool ipgre_netlink_encap_parms(struct nlattr *data[],
641 				      struct ip_tunnel_encap *ipencap)
642 {
643 	bool ret = false;
644 
645 	memset(ipencap, 0, sizeof(*ipencap));
646 
647 	if (!data)
648 		return ret;
649 
650 	if (data[IFLA_GRE_ENCAP_TYPE]) {
651 		ret = true;
652 		ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
653 	}
654 
655 	if (data[IFLA_GRE_ENCAP_FLAGS]) {
656 		ret = true;
657 		ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
658 	}
659 
660 	if (data[IFLA_GRE_ENCAP_SPORT]) {
661 		ret = true;
662 		ipencap->sport = nla_get_u16(data[IFLA_GRE_ENCAP_SPORT]);
663 	}
664 
665 	if (data[IFLA_GRE_ENCAP_DPORT]) {
666 		ret = true;
667 		ipencap->dport = nla_get_u16(data[IFLA_GRE_ENCAP_DPORT]);
668 	}
669 
670 	return ret;
671 }
672 
673 static int gre_tap_init(struct net_device *dev)
674 {
675 	__gre_tunnel_init(dev);
676 
677 	return ip_tunnel_init(dev);
678 }
679 
680 static const struct net_device_ops gre_tap_netdev_ops = {
681 	.ndo_init		= gre_tap_init,
682 	.ndo_uninit		= ip_tunnel_uninit,
683 	.ndo_start_xmit		= gre_tap_xmit,
684 	.ndo_set_mac_address 	= eth_mac_addr,
685 	.ndo_validate_addr	= eth_validate_addr,
686 	.ndo_change_mtu		= ip_tunnel_change_mtu,
687 	.ndo_get_stats64	= ip_tunnel_get_stats64,
688 };
689 
690 static void ipgre_tap_setup(struct net_device *dev)
691 {
692 	ether_setup(dev);
693 	dev->netdev_ops		= &gre_tap_netdev_ops;
694 	dev->priv_flags 	|= IFF_LIVE_ADDR_CHANGE;
695 	ip_tunnel_setup(dev, gre_tap_net_id);
696 }
697 
698 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
699 			 struct nlattr *tb[], struct nlattr *data[])
700 {
701 	struct ip_tunnel_parm p;
702 	struct ip_tunnel_encap ipencap;
703 
704 	if (ipgre_netlink_encap_parms(data, &ipencap)) {
705 		struct ip_tunnel *t = netdev_priv(dev);
706 		int err = ip_tunnel_encap_setup(t, &ipencap);
707 
708 		if (err < 0)
709 			return err;
710 	}
711 
712 	ipgre_netlink_parms(data, tb, &p);
713 	return ip_tunnel_newlink(dev, tb, &p);
714 }
715 
716 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
717 			    struct nlattr *data[])
718 {
719 	struct ip_tunnel_parm p;
720 	struct ip_tunnel_encap ipencap;
721 
722 	if (ipgre_netlink_encap_parms(data, &ipencap)) {
723 		struct ip_tunnel *t = netdev_priv(dev);
724 		int err = ip_tunnel_encap_setup(t, &ipencap);
725 
726 		if (err < 0)
727 			return err;
728 	}
729 
730 	ipgre_netlink_parms(data, tb, &p);
731 	return ip_tunnel_changelink(dev, tb, &p);
732 }
733 
734 static size_t ipgre_get_size(const struct net_device *dev)
735 {
736 	return
737 		/* IFLA_GRE_LINK */
738 		nla_total_size(4) +
739 		/* IFLA_GRE_IFLAGS */
740 		nla_total_size(2) +
741 		/* IFLA_GRE_OFLAGS */
742 		nla_total_size(2) +
743 		/* IFLA_GRE_IKEY */
744 		nla_total_size(4) +
745 		/* IFLA_GRE_OKEY */
746 		nla_total_size(4) +
747 		/* IFLA_GRE_LOCAL */
748 		nla_total_size(4) +
749 		/* IFLA_GRE_REMOTE */
750 		nla_total_size(4) +
751 		/* IFLA_GRE_TTL */
752 		nla_total_size(1) +
753 		/* IFLA_GRE_TOS */
754 		nla_total_size(1) +
755 		/* IFLA_GRE_PMTUDISC */
756 		nla_total_size(1) +
757 		/* IFLA_GRE_ENCAP_TYPE */
758 		nla_total_size(2) +
759 		/* IFLA_GRE_ENCAP_FLAGS */
760 		nla_total_size(2) +
761 		/* IFLA_GRE_ENCAP_SPORT */
762 		nla_total_size(2) +
763 		/* IFLA_GRE_ENCAP_DPORT */
764 		nla_total_size(2) +
765 		0;
766 }
767 
768 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
769 {
770 	struct ip_tunnel *t = netdev_priv(dev);
771 	struct ip_tunnel_parm *p = &t->parms;
772 
773 	if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
774 	    nla_put_be16(skb, IFLA_GRE_IFLAGS, tnl_flags_to_gre_flags(p->i_flags)) ||
775 	    nla_put_be16(skb, IFLA_GRE_OFLAGS, tnl_flags_to_gre_flags(p->o_flags)) ||
776 	    nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
777 	    nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
778 	    nla_put_be32(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
779 	    nla_put_be32(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
780 	    nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
781 	    nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
782 	    nla_put_u8(skb, IFLA_GRE_PMTUDISC,
783 		       !!(p->iph.frag_off & htons(IP_DF))))
784 		goto nla_put_failure;
785 
786 	if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
787 			t->encap.type) ||
788 	    nla_put_u16(skb, IFLA_GRE_ENCAP_SPORT,
789 			t->encap.sport) ||
790 	    nla_put_u16(skb, IFLA_GRE_ENCAP_DPORT,
791 			t->encap.dport) ||
792 	    nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
793 			t->encap.flags))
794 		goto nla_put_failure;
795 
796 	return 0;
797 
798 nla_put_failure:
799 	return -EMSGSIZE;
800 }
801 
802 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
803 	[IFLA_GRE_LINK]		= { .type = NLA_U32 },
804 	[IFLA_GRE_IFLAGS]	= { .type = NLA_U16 },
805 	[IFLA_GRE_OFLAGS]	= { .type = NLA_U16 },
806 	[IFLA_GRE_IKEY]		= { .type = NLA_U32 },
807 	[IFLA_GRE_OKEY]		= { .type = NLA_U32 },
808 	[IFLA_GRE_LOCAL]	= { .len = FIELD_SIZEOF(struct iphdr, saddr) },
809 	[IFLA_GRE_REMOTE]	= { .len = FIELD_SIZEOF(struct iphdr, daddr) },
810 	[IFLA_GRE_TTL]		= { .type = NLA_U8 },
811 	[IFLA_GRE_TOS]		= { .type = NLA_U8 },
812 	[IFLA_GRE_PMTUDISC]	= { .type = NLA_U8 },
813 	[IFLA_GRE_ENCAP_TYPE]	= { .type = NLA_U16 },
814 	[IFLA_GRE_ENCAP_FLAGS]	= { .type = NLA_U16 },
815 	[IFLA_GRE_ENCAP_SPORT]	= { .type = NLA_U16 },
816 	[IFLA_GRE_ENCAP_DPORT]	= { .type = NLA_U16 },
817 };
818 
819 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
820 	.kind		= "gre",
821 	.maxtype	= IFLA_GRE_MAX,
822 	.policy		= ipgre_policy,
823 	.priv_size	= sizeof(struct ip_tunnel),
824 	.setup		= ipgre_tunnel_setup,
825 	.validate	= ipgre_tunnel_validate,
826 	.newlink	= ipgre_newlink,
827 	.changelink	= ipgre_changelink,
828 	.dellink	= ip_tunnel_dellink,
829 	.get_size	= ipgre_get_size,
830 	.fill_info	= ipgre_fill_info,
831 };
832 
833 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
834 	.kind		= "gretap",
835 	.maxtype	= IFLA_GRE_MAX,
836 	.policy		= ipgre_policy,
837 	.priv_size	= sizeof(struct ip_tunnel),
838 	.setup		= ipgre_tap_setup,
839 	.validate	= ipgre_tap_validate,
840 	.newlink	= ipgre_newlink,
841 	.changelink	= ipgre_changelink,
842 	.dellink	= ip_tunnel_dellink,
843 	.get_size	= ipgre_get_size,
844 	.fill_info	= ipgre_fill_info,
845 };
846 
847 static int __net_init ipgre_tap_init_net(struct net *net)
848 {
849 	return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, NULL);
850 }
851 
852 static void __net_exit ipgre_tap_exit_net(struct net *net)
853 {
854 	struct ip_tunnel_net *itn = net_generic(net, gre_tap_net_id);
855 	ip_tunnel_delete_net(itn, &ipgre_tap_ops);
856 }
857 
858 static struct pernet_operations ipgre_tap_net_ops = {
859 	.init = ipgre_tap_init_net,
860 	.exit = ipgre_tap_exit_net,
861 	.id   = &gre_tap_net_id,
862 	.size = sizeof(struct ip_tunnel_net),
863 };
864 
865 static int __init ipgre_init(void)
866 {
867 	int err;
868 
869 	pr_info("GRE over IPv4 tunneling driver\n");
870 
871 	err = register_pernet_device(&ipgre_net_ops);
872 	if (err < 0)
873 		return err;
874 
875 	err = register_pernet_device(&ipgre_tap_net_ops);
876 	if (err < 0)
877 		goto pnet_tap_faied;
878 
879 	err = gre_cisco_register(&ipgre_protocol);
880 	if (err < 0) {
881 		pr_info("%s: can't add protocol\n", __func__);
882 		goto add_proto_failed;
883 	}
884 
885 	err = rtnl_link_register(&ipgre_link_ops);
886 	if (err < 0)
887 		goto rtnl_link_failed;
888 
889 	err = rtnl_link_register(&ipgre_tap_ops);
890 	if (err < 0)
891 		goto tap_ops_failed;
892 
893 	return 0;
894 
895 tap_ops_failed:
896 	rtnl_link_unregister(&ipgre_link_ops);
897 rtnl_link_failed:
898 	gre_cisco_unregister(&ipgre_protocol);
899 add_proto_failed:
900 	unregister_pernet_device(&ipgre_tap_net_ops);
901 pnet_tap_faied:
902 	unregister_pernet_device(&ipgre_net_ops);
903 	return err;
904 }
905 
906 static void __exit ipgre_fini(void)
907 {
908 	rtnl_link_unregister(&ipgre_tap_ops);
909 	rtnl_link_unregister(&ipgre_link_ops);
910 	gre_cisco_unregister(&ipgre_protocol);
911 	unregister_pernet_device(&ipgre_tap_net_ops);
912 	unregister_pernet_device(&ipgre_net_ops);
913 }
914 
915 module_init(ipgre_init);
916 module_exit(ipgre_fini);
917 MODULE_LICENSE("GPL");
918 MODULE_ALIAS_RTNL_LINK("gre");
919 MODULE_ALIAS_RTNL_LINK("gretap");
920 MODULE_ALIAS_NETDEV("gre0");
921 MODULE_ALIAS_NETDEV("gretap0");
922