2 * Linux NET3: GRE over IP protocol decoder.
4 * Authors: Alexey Kuznetsov (kuznet@ms2.inr.ac.ru)
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.
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
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>
39 #include <net/protocol.h>
40 #include <net/ip_tunnels.h>
42 #include <net/checksum.h>
43 #include <net/dsfield.h>
44 #include <net/inet_ecn.h>
46 #include <net/net_namespace.h>
47 #include <net/netns/generic.h>
48 #include <net/rtnetlink.h>
50 #include <net/dst_metadata.h>
56 1. The most important issue is detecting local dead loops.
57 They would cause complete host lockup in transmit, which
58 would be "resolved" by stack overflow or, if queueing is enabled,
59 with infinite looping in net_bh.
61 We cannot track such dead loops during route installation,
62 it is infeasible task. The most general solutions would be
63 to keep skb->encapsulation counter (sort of local ttl),
64 and silently drop packet when it expires. It is a good
65 solution, but it supposes maintaining new variable in ALL
66 skb, even if no tunneling is used.
68 Current solution: xmit_recursion breaks dead loops. This is a percpu
69 counter, since when we enter the first ndo_xmit(), cpu migration is
70 forbidden. We force an exit if this counter reaches RECURSION_LIMIT
72 2. Networking dead loops would not kill routers, but would really
73 kill network. IP hop limit plays role of "t->recursion" in this case,
74 if we copy it from packet being encapsulated to upper header.
75 It is very good solution, but it introduces two problems:
77 - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
78 do not work over tunnels.
79 - traceroute does not work. I planned to relay ICMP from tunnel,
80 so that this problem would be solved and traceroute output
81 would even more informative. This idea appeared to be wrong:
82 only Linux complies to rfc1812 now (yes, guys, Linux is the only
83 true router now :-)), all routers (at least, in neighbourhood of mine)
84 return only 8 bytes of payload. It is the end.
86 Hence, if we want that OSPF worked or traceroute said something reasonable,
87 we should search for another solution.
89 One of them is to parse packet trying to detect inner encapsulation
90 made by our node. It is difficult or even impossible, especially,
91 taking into account fragmentation. TO be short, ttl is not solution at all.
93 Current solution: The solution was UNEXPECTEDLY SIMPLE.
94 We force DF flag on tunnels with preconfigured hop limit,
95 that is ALL. :-) Well, it does not remove the problem completely,
96 but exponential growth of network traffic is changed to linear
97 (branches, that exceed pmtu are pruned) and tunnel mtu
98 rapidly degrades to value <68, where looping stops.
99 Yes, it is not good if there exists a router in the loop,
100 which does not force DF, even when encapsulating packets have DF set.
101 But it is not our problem! Nobody could accuse us, we made
102 all that we could make. Even if it is your gated who injected
103 fatal route to network, even if it were you who configured
104 fatal static route: you are innocent. :-)
109 static bool log_ecn_error = true;
110 module_param(log_ecn_error, bool, 0644);
111 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
113 static struct rtnl_link_ops ipgre_link_ops __read_mostly;
114 static int ipgre_tunnel_init(struct net_device *dev);
116 static int ipgre_net_id __read_mostly;
117 static int gre_tap_net_id __read_mostly;
119 static void ipgre_err(struct sk_buff *skb, u32 info,
120 const struct tnl_ptk_info *tpi)
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.
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
132 Well, I wonder, rfc1812 was written by Cisco employee,
133 what the hell these idiots break standards established
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;
146 case ICMP_PARAMETERPROB:
149 case ICMP_DEST_UNREACH:
152 case ICMP_PORT_UNREACH:
153 /* Impossible event. */
156 /* All others are translated to HOST_UNREACH.
157 rfc2003 contains "deep thoughts" about NET_UNREACH,
158 I believe they are just ether pollution. --ANK
164 case ICMP_TIME_EXCEEDED:
165 if (code != ICMP_EXC_TTL)
167 data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */
174 if (tpi->proto == htons(ETH_P_TEB))
175 itn = net_generic(net, gre_tap_net_id);
177 itn = net_generic(net, ipgre_net_id);
179 iph = (const struct iphdr *)(icmp_hdr(skb) + 1);
180 t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
181 iph->daddr, iph->saddr, tpi->key);
186 #if IS_ENABLED(CONFIG_IPV6)
187 if (tpi->proto == htons(ETH_P_IPV6) &&
188 !ip6_err_gen_icmpv6_unreach(skb, iph->ihl * 4 + tpi->hdr_len,
193 if (t->parms.iph.daddr == 0 ||
194 ipv4_is_multicast(t->parms.iph.daddr))
197 if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
200 if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
204 t->err_time = jiffies;
207 static void gre_err(struct sk_buff *skb, u32 info)
209 /* All the routers (except for Linux) return only
210 * 8 bytes of packet payload. It means, that precise relaying of
211 * ICMP in the real Internet is absolutely infeasible.
213 * Moreover, Cisco "wise men" put GRE key to the third word
214 * in GRE header. It makes impossible maintaining even soft
216 * GRE tunnels with enabled checksum. Tell them "thank you".
218 * Well, I wonder, rfc1812 was written by Cisco employee,
219 * what the hell these idiots break standards established
223 const struct iphdr *iph = (struct iphdr *)skb->data;
224 const int type = icmp_hdr(skb)->type;
225 const int code = icmp_hdr(skb)->code;
226 struct tnl_ptk_info tpi;
227 bool csum_err = false;
229 if (gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP),
231 if (!csum_err) /* ignore csum errors. */
235 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
236 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
237 skb->dev->ifindex, 0, IPPROTO_GRE, 0);
240 if (type == ICMP_REDIRECT) {
241 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex, 0,
246 ipgre_err(skb, info, &tpi);
249 static __be64 key_to_tunnel_id(__be32 key)
252 return (__force __be64)((__force u32)key);
254 return (__force __be64)((__force u64)key << 32);
258 /* Returns the least-significant 32 bits of a __be64. */
259 static __be32 tunnel_id_to_key(__be64 x)
262 return (__force __be32)x;
264 return (__force __be32)((__force u64)x >> 32);
268 static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
269 struct ip_tunnel_net *itn, int hdr_len, bool raw_proto)
271 struct metadata_dst *tun_dst = NULL;
272 const struct iphdr *iph;
273 struct ip_tunnel *tunnel;
276 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
277 iph->saddr, iph->daddr, tpi->key);
280 if (__iptunnel_pull_header(skb, hdr_len, tpi->proto,
281 raw_proto, false) < 0)
284 if (tunnel->dev->type != ARPHRD_NONE)
285 skb_pop_mac_header(skb);
287 skb_reset_mac_header(skb);
288 if (tunnel->collect_md) {
292 flags = tpi->flags & (TUNNEL_CSUM | TUNNEL_KEY);
293 tun_id = key_to_tunnel_id(tpi->key);
294 tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
296 return PACKET_REJECT;
299 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
309 static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
312 struct net *net = dev_net(skb->dev);
313 struct ip_tunnel_net *itn;
316 if (tpi->proto == htons(ETH_P_TEB))
317 itn = net_generic(net, gre_tap_net_id);
319 itn = net_generic(net, ipgre_net_id);
321 res = __ipgre_rcv(skb, tpi, itn, hdr_len, false);
322 if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) {
323 /* ipgre tunnels in collect metadata mode should receive
324 * also ETH_P_TEB traffic.
326 itn = net_generic(net, ipgre_net_id);
327 res = __ipgre_rcv(skb, tpi, itn, hdr_len, true);
332 static int gre_rcv(struct sk_buff *skb)
334 struct tnl_ptk_info tpi;
335 bool csum_err = false;
338 #ifdef CONFIG_NET_IPGRE_BROADCAST
339 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
340 /* Looped back packet, drop it! */
341 if (rt_is_output_route(skb_rtable(skb)))
346 hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0);
350 if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
353 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
359 static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
360 const struct iphdr *tnl_params,
363 struct ip_tunnel *tunnel = netdev_priv(dev);
365 if (tunnel->parms.o_flags & TUNNEL_SEQ)
368 /* Push GRE header. */
369 gre_build_header(skb, tunnel->tun_hlen,
370 tunnel->parms.o_flags, proto, tunnel->parms.o_key,
371 htonl(tunnel->o_seqno));
373 skb_set_inner_protocol(skb, proto);
374 ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
377 static int gre_handle_offloads(struct sk_buff *skb, bool csum)
379 return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
382 static struct rtable *gre_get_rt(struct sk_buff *skb,
383 struct net_device *dev,
385 const struct ip_tunnel_key *key)
387 struct net *net = dev_net(dev);
389 memset(fl, 0, sizeof(*fl));
390 fl->daddr = key->u.ipv4.dst;
391 fl->saddr = key->u.ipv4.src;
392 fl->flowi4_tos = RT_TOS(key->tos);
393 fl->flowi4_mark = skb->mark;
394 fl->flowi4_proto = IPPROTO_GRE;
396 return ip_route_output_key(net, fl);
399 static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev,
402 struct ip_tunnel_info *tun_info;
403 const struct ip_tunnel_key *key;
404 struct rtable *rt = NULL;
412 tun_info = skb_tunnel_info(skb);
413 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
414 ip_tunnel_info_af(tun_info) != AF_INET))
417 key = &tun_info->key;
418 use_cache = ip_tunnel_dst_cache_usable(skb, tun_info);
420 rt = dst_cache_get_ip4(&tun_info->dst_cache, &fl.saddr);
422 rt = gre_get_rt(skb, dev, &fl, key);
426 dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst,
430 tunnel_hlen = gre_calc_hlen(key->tun_flags);
432 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
433 + tunnel_hlen + sizeof(struct iphdr);
434 if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
435 int head_delta = SKB_DATA_ALIGN(min_headroom -
438 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
444 /* Push Tunnel header. */
445 if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM)))
448 flags = tun_info->key.tun_flags & (TUNNEL_CSUM | TUNNEL_KEY);
449 gre_build_header(skb, tunnel_hlen, flags, proto,
450 tunnel_id_to_key(tun_info->key.tun_id), 0);
452 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
454 iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
455 key->tos, key->ttl, df, false);
462 dev->stats.tx_dropped++;
465 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
467 struct ip_tunnel_info *info = skb_tunnel_info(skb);
471 if (ip_tunnel_info_af(info) != AF_INET)
474 rt = gre_get_rt(skb, dev, &fl4, &info->key);
479 info->key.u.ipv4.src = fl4.saddr;
483 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
484 struct net_device *dev)
486 struct ip_tunnel *tunnel = netdev_priv(dev);
487 const struct iphdr *tnl_params;
489 if (tunnel->collect_md) {
490 gre_fb_xmit(skb, dev, skb->protocol);
494 if (dev->header_ops) {
495 /* Need space for new headers */
496 if (skb_cow_head(skb, dev->needed_headroom -
497 (tunnel->hlen + sizeof(struct iphdr))))
500 tnl_params = (const struct iphdr *)skb->data;
502 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing
505 skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
506 skb_reset_mac_header(skb);
508 if (skb_cow_head(skb, dev->needed_headroom))
511 tnl_params = &tunnel->parms.iph;
514 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
517 __gre_xmit(skb, dev, tnl_params, skb->protocol);
522 dev->stats.tx_dropped++;
526 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
527 struct net_device *dev)
529 struct ip_tunnel *tunnel = netdev_priv(dev);
531 if (tunnel->collect_md) {
532 gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
536 if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
539 if (skb_cow_head(skb, dev->needed_headroom))
542 __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
547 dev->stats.tx_dropped++;
551 static int ipgre_tunnel_ioctl(struct net_device *dev,
552 struct ifreq *ifr, int cmd)
555 struct ip_tunnel_parm p;
557 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
559 if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
560 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
561 p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) ||
562 ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING)))
565 p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
566 p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
568 err = ip_tunnel_ioctl(dev, &p, cmd);
572 p.i_flags = gre_tnl_flags_to_gre_flags(p.i_flags);
573 p.o_flags = gre_tnl_flags_to_gre_flags(p.o_flags);
575 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
580 /* Nice toy. Unfortunately, useless in real life :-)
581 It allows to construct virtual multiprotocol broadcast "LAN"
582 over the Internet, provided multicast routing is tuned.
585 I have no idea was this bicycle invented before me,
586 so that I had to set ARPHRD_IPGRE to a random value.
587 I have an impression, that Cisco could make something similar,
588 but this feature is apparently missing in IOS<=11.2(8).
590 I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
591 with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
593 ping -t 255 224.66.66.66
595 If nobody answers, mbone does not work.
597 ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
598 ip addr add 10.66.66.<somewhat>/24 dev Universe
600 ifconfig Universe add fe80::<Your_real_addr>/10
601 ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
604 ftp fec0:6666:6666::193.233.7.65
607 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
609 const void *daddr, const void *saddr, unsigned int len)
611 struct ip_tunnel *t = netdev_priv(dev);
613 struct gre_base_hdr *greh;
615 iph = (struct iphdr *)skb_push(skb, t->hlen + sizeof(*iph));
616 greh = (struct gre_base_hdr *)(iph+1);
617 greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
618 greh->protocol = htons(type);
620 memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
622 /* Set the source hardware address. */
624 memcpy(&iph->saddr, saddr, 4);
626 memcpy(&iph->daddr, daddr, 4);
628 return t->hlen + sizeof(*iph);
630 return -(t->hlen + sizeof(*iph));
633 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
635 const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
636 memcpy(haddr, &iph->saddr, 4);
640 static const struct header_ops ipgre_header_ops = {
641 .create = ipgre_header,
642 .parse = ipgre_header_parse,
645 #ifdef CONFIG_NET_IPGRE_BROADCAST
646 static int ipgre_open(struct net_device *dev)
648 struct ip_tunnel *t = netdev_priv(dev);
650 if (ipv4_is_multicast(t->parms.iph.daddr)) {
654 rt = ip_route_output_gre(t->net, &fl4,
658 RT_TOS(t->parms.iph.tos),
661 return -EADDRNOTAVAIL;
664 if (!__in_dev_get_rtnl(dev))
665 return -EADDRNOTAVAIL;
666 t->mlink = dev->ifindex;
667 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
672 static int ipgre_close(struct net_device *dev)
674 struct ip_tunnel *t = netdev_priv(dev);
676 if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
677 struct in_device *in_dev;
678 in_dev = inetdev_by_index(t->net, t->mlink);
680 ip_mc_dec_group(in_dev, t->parms.iph.daddr);
686 static const struct net_device_ops ipgre_netdev_ops = {
687 .ndo_init = ipgre_tunnel_init,
688 .ndo_uninit = ip_tunnel_uninit,
689 #ifdef CONFIG_NET_IPGRE_BROADCAST
690 .ndo_open = ipgre_open,
691 .ndo_stop = ipgre_close,
693 .ndo_start_xmit = ipgre_xmit,
694 .ndo_do_ioctl = ipgre_tunnel_ioctl,
695 .ndo_change_mtu = ip_tunnel_change_mtu,
696 .ndo_get_stats64 = ip_tunnel_get_stats64,
697 .ndo_get_iflink = ip_tunnel_get_iflink,
700 #define GRE_FEATURES (NETIF_F_SG | \
705 static void ipgre_tunnel_setup(struct net_device *dev)
707 dev->netdev_ops = &ipgre_netdev_ops;
708 dev->type = ARPHRD_IPGRE;
709 ip_tunnel_setup(dev, ipgre_net_id);
712 static void __gre_tunnel_init(struct net_device *dev)
714 struct ip_tunnel *tunnel;
717 tunnel = netdev_priv(dev);
718 tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
719 tunnel->parms.iph.protocol = IPPROTO_GRE;
721 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
723 t_hlen = tunnel->hlen + sizeof(struct iphdr);
725 dev->needed_headroom = LL_MAX_HEADER + t_hlen + 4;
726 dev->mtu = ETH_DATA_LEN - t_hlen - 4;
728 dev->features |= GRE_FEATURES;
729 dev->hw_features |= GRE_FEATURES;
731 if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
732 /* TCP offload with GRE SEQ is not supported, nor
733 * can we support 2 levels of outer headers requiring
736 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
737 (tunnel->encap.type == TUNNEL_ENCAP_NONE)) {
738 dev->features |= NETIF_F_GSO_SOFTWARE;
739 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
742 /* Can use a lockless transmit, unless we generate
745 dev->features |= NETIF_F_LLTX;
749 static int ipgre_tunnel_init(struct net_device *dev)
751 struct ip_tunnel *tunnel = netdev_priv(dev);
752 struct iphdr *iph = &tunnel->parms.iph;
754 __gre_tunnel_init(dev);
756 memcpy(dev->dev_addr, &iph->saddr, 4);
757 memcpy(dev->broadcast, &iph->daddr, 4);
759 dev->flags = IFF_NOARP;
763 if (iph->daddr && !tunnel->collect_md) {
764 #ifdef CONFIG_NET_IPGRE_BROADCAST
765 if (ipv4_is_multicast(iph->daddr)) {
768 dev->flags = IFF_BROADCAST;
769 dev->header_ops = &ipgre_header_ops;
772 } else if (!tunnel->collect_md) {
773 dev->header_ops = &ipgre_header_ops;
776 return ip_tunnel_init(dev);
779 static const struct gre_protocol ipgre_protocol = {
781 .err_handler = gre_err,
784 static int __net_init ipgre_init_net(struct net *net)
786 return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
789 static void __net_exit ipgre_exit_net(struct net *net)
791 struct ip_tunnel_net *itn = net_generic(net, ipgre_net_id);
792 ip_tunnel_delete_net(itn, &ipgre_link_ops);
795 static struct pernet_operations ipgre_net_ops = {
796 .init = ipgre_init_net,
797 .exit = ipgre_exit_net,
799 .size = sizeof(struct ip_tunnel_net),
802 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
810 if (data[IFLA_GRE_IFLAGS])
811 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
812 if (data[IFLA_GRE_OFLAGS])
813 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
814 if (flags & (GRE_VERSION|GRE_ROUTING))
817 if (data[IFLA_GRE_COLLECT_METADATA] &&
818 data[IFLA_GRE_ENCAP_TYPE] &&
819 nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
825 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
829 if (tb[IFLA_ADDRESS]) {
830 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
832 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
833 return -EADDRNOTAVAIL;
839 if (data[IFLA_GRE_REMOTE]) {
840 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
846 return ipgre_tunnel_validate(tb, data);
849 static int ipgre_netlink_parms(struct net_device *dev,
850 struct nlattr *data[],
852 struct ip_tunnel_parm *parms)
854 struct ip_tunnel *t = netdev_priv(dev);
856 memset(parms, 0, sizeof(*parms));
858 parms->iph.protocol = IPPROTO_GRE;
863 if (data[IFLA_GRE_LINK])
864 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
866 if (data[IFLA_GRE_IFLAGS])
867 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
869 if (data[IFLA_GRE_OFLAGS])
870 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
872 if (data[IFLA_GRE_IKEY])
873 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
875 if (data[IFLA_GRE_OKEY])
876 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
878 if (data[IFLA_GRE_LOCAL])
879 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
881 if (data[IFLA_GRE_REMOTE])
882 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
884 if (data[IFLA_GRE_TTL])
885 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
887 if (data[IFLA_GRE_TOS])
888 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
890 if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
893 parms->iph.frag_off = htons(IP_DF);
896 if (data[IFLA_GRE_COLLECT_METADATA]) {
897 t->collect_md = true;
898 if (dev->type == ARPHRD_IPGRE)
899 dev->type = ARPHRD_NONE;
902 if (data[IFLA_GRE_IGNORE_DF]) {
903 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
904 && (parms->iph.frag_off & htons(IP_DF)))
906 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
912 /* This function returns true when ENCAP attributes are present in the nl msg */
913 static bool ipgre_netlink_encap_parms(struct nlattr *data[],
914 struct ip_tunnel_encap *ipencap)
918 memset(ipencap, 0, sizeof(*ipencap));
923 if (data[IFLA_GRE_ENCAP_TYPE]) {
925 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
928 if (data[IFLA_GRE_ENCAP_FLAGS]) {
930 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
933 if (data[IFLA_GRE_ENCAP_SPORT]) {
935 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
938 if (data[IFLA_GRE_ENCAP_DPORT]) {
940 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
946 static int gre_tap_init(struct net_device *dev)
948 __gre_tunnel_init(dev);
949 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
951 return ip_tunnel_init(dev);
954 static const struct net_device_ops gre_tap_netdev_ops = {
955 .ndo_init = gre_tap_init,
956 .ndo_uninit = ip_tunnel_uninit,
957 .ndo_start_xmit = gre_tap_xmit,
958 .ndo_set_mac_address = eth_mac_addr,
959 .ndo_validate_addr = eth_validate_addr,
960 .ndo_change_mtu = ip_tunnel_change_mtu,
961 .ndo_get_stats64 = ip_tunnel_get_stats64,
962 .ndo_get_iflink = ip_tunnel_get_iflink,
963 .ndo_fill_metadata_dst = gre_fill_metadata_dst,
966 static void ipgre_tap_setup(struct net_device *dev)
969 dev->netdev_ops = &gre_tap_netdev_ops;
970 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
971 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
972 ip_tunnel_setup(dev, gre_tap_net_id);
975 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
976 struct nlattr *tb[], struct nlattr *data[])
978 struct ip_tunnel_parm p;
979 struct ip_tunnel_encap ipencap;
982 if (ipgre_netlink_encap_parms(data, &ipencap)) {
983 struct ip_tunnel *t = netdev_priv(dev);
984 err = ip_tunnel_encap_setup(t, &ipencap);
990 err = ipgre_netlink_parms(dev, data, tb, &p);
993 return ip_tunnel_newlink(dev, tb, &p);
996 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
997 struct nlattr *data[])
999 struct ip_tunnel_parm p;
1000 struct ip_tunnel_encap ipencap;
1003 if (ipgre_netlink_encap_parms(data, &ipencap)) {
1004 struct ip_tunnel *t = netdev_priv(dev);
1005 err = ip_tunnel_encap_setup(t, &ipencap);
1011 err = ipgre_netlink_parms(dev, data, tb, &p);
1014 return ip_tunnel_changelink(dev, tb, &p);
1017 static size_t ipgre_get_size(const struct net_device *dev)
1022 /* IFLA_GRE_IFLAGS */
1024 /* IFLA_GRE_OFLAGS */
1030 /* IFLA_GRE_LOCAL */
1032 /* IFLA_GRE_REMOTE */
1038 /* IFLA_GRE_PMTUDISC */
1040 /* IFLA_GRE_ENCAP_TYPE */
1042 /* IFLA_GRE_ENCAP_FLAGS */
1044 /* IFLA_GRE_ENCAP_SPORT */
1046 /* IFLA_GRE_ENCAP_DPORT */
1048 /* IFLA_GRE_COLLECT_METADATA */
1050 /* IFLA_GRE_IGNORE_DF */
1055 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1057 struct ip_tunnel *t = netdev_priv(dev);
1058 struct ip_tunnel_parm *p = &t->parms;
1060 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1061 nla_put_be16(skb, IFLA_GRE_IFLAGS,
1062 gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1063 nla_put_be16(skb, IFLA_GRE_OFLAGS,
1064 gre_tnl_flags_to_gre_flags(p->o_flags)) ||
1065 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1066 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1067 nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1068 nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1069 nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1070 nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1071 nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1072 !!(p->iph.frag_off & htons(IP_DF))))
1073 goto nla_put_failure;
1075 if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1077 nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1079 nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1081 nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1083 goto nla_put_failure;
1085 if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1086 goto nla_put_failure;
1088 if (t->collect_md) {
1089 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1090 goto nla_put_failure;
1099 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1100 [IFLA_GRE_LINK] = { .type = NLA_U32 },
1101 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 },
1102 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 },
1103 [IFLA_GRE_IKEY] = { .type = NLA_U32 },
1104 [IFLA_GRE_OKEY] = { .type = NLA_U32 },
1105 [IFLA_GRE_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1106 [IFLA_GRE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1107 [IFLA_GRE_TTL] = { .type = NLA_U8 },
1108 [IFLA_GRE_TOS] = { .type = NLA_U8 },
1109 [IFLA_GRE_PMTUDISC] = { .type = NLA_U8 },
1110 [IFLA_GRE_ENCAP_TYPE] = { .type = NLA_U16 },
1111 [IFLA_GRE_ENCAP_FLAGS] = { .type = NLA_U16 },
1112 [IFLA_GRE_ENCAP_SPORT] = { .type = NLA_U16 },
1113 [IFLA_GRE_ENCAP_DPORT] = { .type = NLA_U16 },
1114 [IFLA_GRE_COLLECT_METADATA] = { .type = NLA_FLAG },
1115 [IFLA_GRE_IGNORE_DF] = { .type = NLA_U8 },
1118 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1120 .maxtype = IFLA_GRE_MAX,
1121 .policy = ipgre_policy,
1122 .priv_size = sizeof(struct ip_tunnel),
1123 .setup = ipgre_tunnel_setup,
1124 .validate = ipgre_tunnel_validate,
1125 .newlink = ipgre_newlink,
1126 .changelink = ipgre_changelink,
1127 .dellink = ip_tunnel_dellink,
1128 .get_size = ipgre_get_size,
1129 .fill_info = ipgre_fill_info,
1130 .get_link_net = ip_tunnel_get_link_net,
1133 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1135 .maxtype = IFLA_GRE_MAX,
1136 .policy = ipgre_policy,
1137 .priv_size = sizeof(struct ip_tunnel),
1138 .setup = ipgre_tap_setup,
1139 .validate = ipgre_tap_validate,
1140 .newlink = ipgre_newlink,
1141 .changelink = ipgre_changelink,
1142 .dellink = ip_tunnel_dellink,
1143 .get_size = ipgre_get_size,
1144 .fill_info = ipgre_fill_info,
1145 .get_link_net = ip_tunnel_get_link_net,
1148 struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1149 u8 name_assign_type)
1151 struct nlattr *tb[IFLA_MAX + 1];
1152 struct net_device *dev;
1153 LIST_HEAD(list_kill);
1154 struct ip_tunnel *t;
1157 memset(&tb, 0, sizeof(tb));
1159 dev = rtnl_create_link(net, name, name_assign_type,
1160 &ipgre_tap_ops, tb);
1164 /* Configure flow based GRE device. */
1165 t = netdev_priv(dev);
1166 t->collect_md = true;
1168 err = ipgre_newlink(net, dev, tb, NULL);
1171 return ERR_PTR(err);
1174 /* openvswitch users expect packet sizes to be unrestricted,
1175 * so set the largest MTU we can.
1177 err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1181 err = rtnl_configure_link(dev, NULL);
1187 ip_tunnel_dellink(dev, &list_kill);
1188 unregister_netdevice_many(&list_kill);
1189 return ERR_PTR(err);
1191 EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1193 static int __net_init ipgre_tap_init_net(struct net *net)
1195 return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1198 static void __net_exit ipgre_tap_exit_net(struct net *net)
1200 struct ip_tunnel_net *itn = net_generic(net, gre_tap_net_id);
1201 ip_tunnel_delete_net(itn, &ipgre_tap_ops);
1204 static struct pernet_operations ipgre_tap_net_ops = {
1205 .init = ipgre_tap_init_net,
1206 .exit = ipgre_tap_exit_net,
1207 .id = &gre_tap_net_id,
1208 .size = sizeof(struct ip_tunnel_net),
1211 static int __init ipgre_init(void)
1215 pr_info("GRE over IPv4 tunneling driver\n");
1217 err = register_pernet_device(&ipgre_net_ops);
1221 err = register_pernet_device(&ipgre_tap_net_ops);
1223 goto pnet_tap_faied;
1225 err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1227 pr_info("%s: can't add protocol\n", __func__);
1228 goto add_proto_failed;
1231 err = rtnl_link_register(&ipgre_link_ops);
1233 goto rtnl_link_failed;
1235 err = rtnl_link_register(&ipgre_tap_ops);
1237 goto tap_ops_failed;
1242 rtnl_link_unregister(&ipgre_link_ops);
1244 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1246 unregister_pernet_device(&ipgre_tap_net_ops);
1248 unregister_pernet_device(&ipgre_net_ops);
1252 static void __exit ipgre_fini(void)
1254 rtnl_link_unregister(&ipgre_tap_ops);
1255 rtnl_link_unregister(&ipgre_link_ops);
1256 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1257 unregister_pernet_device(&ipgre_tap_net_ops);
1258 unregister_pernet_device(&ipgre_net_ops);
1261 module_init(ipgre_init);
1262 module_exit(ipgre_fini);
1263 MODULE_LICENSE("GPL");
1264 MODULE_ALIAS_RTNL_LINK("gre");
1265 MODULE_ALIAS_RTNL_LINK("gretap");
1266 MODULE_ALIAS_NETDEV("gre0");
1267 MODULE_ALIAS_NETDEV("gretap0");