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>
52 #if IS_ENABLED(CONFIG_IPV6)
54 #include <net/ip6_fib.h>
55 #include <net/ip6_route.h>
62 1. The most important issue is detecting local dead loops.
63 They would cause complete host lockup in transmit, which
64 would be "resolved" by stack overflow or, if queueing is enabled,
65 with infinite looping in net_bh.
67 We cannot track such dead loops during route installation,
68 it is infeasible task. The most general solutions would be
69 to keep skb->encapsulation counter (sort of local ttl),
70 and silently drop packet when it expires. It is a good
71 solution, but it supposes maintaining new variable in ALL
72 skb, even if no tunneling is used.
74 Current solution: xmit_recursion breaks dead loops. This is a percpu
75 counter, since when we enter the first ndo_xmit(), cpu migration is
76 forbidden. We force an exit if this counter reaches RECURSION_LIMIT
78 2. Networking dead loops would not kill routers, but would really
79 kill network. IP hop limit plays role of "t->recursion" in this case,
80 if we copy it from packet being encapsulated to upper header.
81 It is very good solution, but it introduces two problems:
83 - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
84 do not work over tunnels.
85 - traceroute does not work. I planned to relay ICMP from tunnel,
86 so that this problem would be solved and traceroute output
87 would even more informative. This idea appeared to be wrong:
88 only Linux complies to rfc1812 now (yes, guys, Linux is the only
89 true router now :-)), all routers (at least, in neighbourhood of mine)
90 return only 8 bytes of payload. It is the end.
92 Hence, if we want that OSPF worked or traceroute said something reasonable,
93 we should search for another solution.
95 One of them is to parse packet trying to detect inner encapsulation
96 made by our node. It is difficult or even impossible, especially,
97 taking into account fragmentation. TO be short, ttl is not solution at all.
99 Current solution: The solution was UNEXPECTEDLY SIMPLE.
100 We force DF flag on tunnels with preconfigured hop limit,
101 that is ALL. :-) Well, it does not remove the problem completely,
102 but exponential growth of network traffic is changed to linear
103 (branches, that exceed pmtu are pruned) and tunnel mtu
104 rapidly degrades to value <68, where looping stops.
105 Yes, it is not good if there exists a router in the loop,
106 which does not force DF, even when encapsulating packets have DF set.
107 But it is not our problem! Nobody could accuse us, we made
108 all that we could make. Even if it is your gated who injected
109 fatal route to network, even if it were you who configured
110 fatal static route: you are innocent. :-)
115 static bool log_ecn_error = true;
116 module_param(log_ecn_error, bool, 0644);
117 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
119 static struct rtnl_link_ops ipgre_link_ops __read_mostly;
120 static int ipgre_tunnel_init(struct net_device *dev);
122 static int ipgre_net_id __read_mostly;
123 static int gre_tap_net_id __read_mostly;
125 static int ip_gre_calc_hlen(__be16 o_flags)
129 if (o_flags & TUNNEL_CSUM)
131 if (o_flags & TUNNEL_KEY)
133 if (o_flags & TUNNEL_SEQ)
138 static __be16 gre_flags_to_tnl_flags(__be16 flags)
142 if (flags & GRE_CSUM)
143 tflags |= TUNNEL_CSUM;
144 if (flags & GRE_ROUTING)
145 tflags |= TUNNEL_ROUTING;
147 tflags |= TUNNEL_KEY;
149 tflags |= TUNNEL_SEQ;
150 if (flags & GRE_STRICT)
151 tflags |= TUNNEL_STRICT;
153 tflags |= TUNNEL_REC;
154 if (flags & GRE_VERSION)
155 tflags |= TUNNEL_VERSION;
160 static __be16 tnl_flags_to_gre_flags(__be16 tflags)
164 if (tflags & TUNNEL_CSUM)
166 if (tflags & TUNNEL_ROUTING)
167 flags |= GRE_ROUTING;
168 if (tflags & TUNNEL_KEY)
170 if (tflags & TUNNEL_SEQ)
172 if (tflags & TUNNEL_STRICT)
174 if (tflags & TUNNEL_REC)
176 if (tflags & TUNNEL_VERSION)
177 flags |= GRE_VERSION;
182 static int parse_gre_header(struct sk_buff *skb, struct tnl_ptk_info *tpi,
185 const struct gre_base_hdr *greh;
189 if (unlikely(!pskb_may_pull(skb, sizeof(struct gre_base_hdr))))
192 greh = (struct gre_base_hdr *)skb_transport_header(skb);
193 if (unlikely(greh->flags & (GRE_VERSION | GRE_ROUTING)))
196 tpi->flags = gre_flags_to_tnl_flags(greh->flags);
197 hdr_len = ip_gre_calc_hlen(tpi->flags);
199 if (!pskb_may_pull(skb, hdr_len))
202 greh = (struct gre_base_hdr *)skb_transport_header(skb);
203 tpi->proto = greh->protocol;
205 options = (__be32 *)(greh + 1);
206 if (greh->flags & GRE_CSUM) {
207 if (skb_checksum_simple_validate(skb)) {
212 skb_checksum_try_convert(skb, IPPROTO_GRE, 0,
213 null_compute_pseudo);
217 if (greh->flags & GRE_KEY) {
223 if (unlikely(greh->flags & GRE_SEQ)) {
229 /* WCCP version 1 and 2 protocol decoding.
230 * - Change protocol to IP
231 * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
233 if (greh->flags == 0 && tpi->proto == htons(ETH_P_WCCP)) {
234 tpi->proto = htons(ETH_P_IP);
235 if ((*(u8 *)options & 0xF0) != 0x40) {
237 if (!pskb_may_pull(skb, hdr_len))
241 return iptunnel_pull_header(skb, hdr_len, tpi->proto);
244 static void ipgre_err(struct sk_buff *skb, u32 info,
245 const struct tnl_ptk_info *tpi)
248 /* All the routers (except for Linux) return only
249 8 bytes of packet payload. It means, that precise relaying of
250 ICMP in the real Internet is absolutely infeasible.
252 Moreover, Cisco "wise men" put GRE key to the third word
253 in GRE header. It makes impossible maintaining even soft
254 state for keyed GRE tunnels with enabled checksum. Tell
257 Well, I wonder, rfc1812 was written by Cisco employee,
258 what the hell these idiots break standards established
261 struct net *net = dev_net(skb->dev);
262 struct ip_tunnel_net *itn;
263 const struct iphdr *iph;
264 const int type = icmp_hdr(skb)->type;
265 const int code = icmp_hdr(skb)->code;
270 case ICMP_PARAMETERPROB:
273 case ICMP_DEST_UNREACH:
276 case ICMP_PORT_UNREACH:
277 /* Impossible event. */
280 /* All others are translated to HOST_UNREACH.
281 rfc2003 contains "deep thoughts" about NET_UNREACH,
282 I believe they are just ether pollution. --ANK
288 case ICMP_TIME_EXCEEDED:
289 if (code != ICMP_EXC_TTL)
297 if (tpi->proto == htons(ETH_P_TEB))
298 itn = net_generic(net, gre_tap_net_id);
300 itn = net_generic(net, ipgre_net_id);
302 iph = (const struct iphdr *)(icmp_hdr(skb) + 1);
303 t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
304 iph->daddr, iph->saddr, tpi->key);
309 if (t->parms.iph.daddr == 0 ||
310 ipv4_is_multicast(t->parms.iph.daddr))
313 if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
316 if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
320 t->err_time = jiffies;
323 static void gre_err(struct sk_buff *skb, u32 info)
325 /* All the routers (except for Linux) return only
326 * 8 bytes of packet payload. It means, that precise relaying of
327 * ICMP in the real Internet is absolutely infeasible.
329 * Moreover, Cisco "wise men" put GRE key to the third word
330 * in GRE header. It makes impossible maintaining even soft
332 * GRE tunnels with enabled checksum. Tell them "thank you".
334 * Well, I wonder, rfc1812 was written by Cisco employee,
335 * what the hell these idiots break standards established
339 const int type = icmp_hdr(skb)->type;
340 const int code = icmp_hdr(skb)->code;
341 struct tnl_ptk_info tpi;
342 bool csum_err = false;
344 if (parse_gre_header(skb, &tpi, &csum_err)) {
345 if (!csum_err) /* ignore csum errors. */
349 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
350 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
351 skb->dev->ifindex, 0, IPPROTO_GRE, 0);
354 if (type == ICMP_REDIRECT) {
355 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex, 0,
360 ipgre_err(skb, info, &tpi);
363 static __be64 key_to_tunnel_id(__be32 key)
366 return (__force __be64)((__force u32)key);
368 return (__force __be64)((__force u64)key << 32);
372 /* Returns the least-significant 32 bits of a __be64. */
373 static __be32 tunnel_id_to_key(__be64 x)
376 return (__force __be32)x;
378 return (__force __be32)((__force u64)x >> 32);
382 static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi)
384 struct net *net = dev_net(skb->dev);
385 struct metadata_dst *tun_dst = NULL;
386 struct ip_tunnel_net *itn;
387 const struct iphdr *iph;
388 struct ip_tunnel *tunnel;
390 if (tpi->proto == htons(ETH_P_TEB))
391 itn = net_generic(net, gre_tap_net_id);
393 itn = net_generic(net, ipgre_net_id);
396 tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
397 iph->saddr, iph->daddr, tpi->key);
400 skb_pop_mac_header(skb);
401 if (tunnel->collect_md) {
405 flags = tpi->flags & (TUNNEL_CSUM | TUNNEL_KEY);
406 tun_id = key_to_tunnel_id(tpi->key);
407 tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
409 return PACKET_REJECT;
412 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
415 return PACKET_REJECT;
418 static int gre_rcv(struct sk_buff *skb)
420 struct tnl_ptk_info tpi;
421 bool csum_err = false;
423 #ifdef CONFIG_NET_IPGRE_BROADCAST
424 if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
425 /* Looped back packet, drop it! */
426 if (rt_is_output_route(skb_rtable(skb)))
431 if (parse_gre_header(skb, &tpi, &csum_err) < 0)
434 if (ipgre_rcv(skb, &tpi) == PACKET_RCVD)
437 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
443 static void build_header(struct sk_buff *skb, int hdr_len, __be16 flags,
444 __be16 proto, __be32 key, __be32 seq)
446 struct gre_base_hdr *greh;
448 skb_push(skb, hdr_len);
450 skb_reset_transport_header(skb);
451 greh = (struct gre_base_hdr *)skb->data;
452 greh->flags = tnl_flags_to_gre_flags(flags);
453 greh->protocol = proto;
455 if (flags & (TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_SEQ)) {
456 __be32 *ptr = (__be32 *)(((u8 *)greh) + hdr_len - 4);
458 if (flags & TUNNEL_SEQ) {
462 if (flags & TUNNEL_KEY) {
466 if (flags & TUNNEL_CSUM &&
467 !(skb_shinfo(skb)->gso_type &
468 (SKB_GSO_GRE | SKB_GSO_GRE_CSUM))) {
470 *(__sum16 *)ptr = csum_fold(skb_checksum(skb, 0,
476 static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
477 const struct iphdr *tnl_params,
480 struct ip_tunnel *tunnel = netdev_priv(dev);
482 if (tunnel->parms.o_flags & TUNNEL_SEQ)
485 /* Push GRE header. */
486 build_header(skb, tunnel->tun_hlen, tunnel->parms.o_flags,
487 proto, tunnel->parms.o_key, htonl(tunnel->o_seqno));
489 skb_set_inner_protocol(skb, proto);
490 ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
493 static struct sk_buff *gre_handle_offloads(struct sk_buff *skb,
496 return iptunnel_handle_offloads(skb, csum,
497 csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
500 static struct rtable *gre_get_rt(struct sk_buff *skb,
501 struct net_device *dev,
503 const struct ip_tunnel_key *key)
505 struct net *net = dev_net(dev);
507 memset(fl, 0, sizeof(*fl));
508 fl->daddr = key->u.ipv4.dst;
509 fl->saddr = key->u.ipv4.src;
510 fl->flowi4_tos = RT_TOS(key->tos);
511 fl->flowi4_mark = skb->mark;
512 fl->flowi4_proto = IPPROTO_GRE;
514 return ip_route_output_key(net, fl);
517 static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev)
519 struct ip_tunnel_info *tun_info;
520 const struct ip_tunnel_key *key;
528 tun_info = skb_tunnel_info(skb);
529 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
530 ip_tunnel_info_af(tun_info) != AF_INET))
533 key = &tun_info->key;
534 rt = gre_get_rt(skb, dev, &fl, key);
538 tunnel_hlen = ip_gre_calc_hlen(key->tun_flags);
540 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
541 + tunnel_hlen + sizeof(struct iphdr);
542 if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
543 int head_delta = SKB_DATA_ALIGN(min_headroom -
546 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
552 /* Push Tunnel header. */
553 skb = gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM));
559 flags = tun_info->key.tun_flags & (TUNNEL_CSUM | TUNNEL_KEY);
560 build_header(skb, tunnel_hlen, flags, htons(ETH_P_TEB),
561 tunnel_id_to_key(tun_info->key.tun_id), 0);
563 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
565 iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
566 key->tos, key->ttl, df, false);
573 dev->stats.tx_dropped++;
576 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
578 struct ip_tunnel_info *info = skb_tunnel_info(skb);
582 if (ip_tunnel_info_af(info) != AF_INET)
585 rt = gre_get_rt(skb, dev, &fl4, &info->key);
590 info->key.u.ipv4.src = fl4.saddr;
594 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
595 struct net_device *dev)
597 struct ip_tunnel *tunnel = netdev_priv(dev);
598 const struct iphdr *tnl_params;
600 if (tunnel->collect_md) {
601 gre_fb_xmit(skb, dev);
605 if (dev->header_ops) {
606 /* Need space for new headers */
607 if (skb_cow_head(skb, dev->needed_headroom -
608 (tunnel->hlen + sizeof(struct iphdr))))
611 tnl_params = (const struct iphdr *)skb->data;
613 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing
616 skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
617 skb_reset_mac_header(skb);
619 if (skb_cow_head(skb, dev->needed_headroom))
622 tnl_params = &tunnel->parms.iph;
625 skb = gre_handle_offloads(skb, !!(tunnel->parms.o_flags&TUNNEL_CSUM));
629 __gre_xmit(skb, dev, tnl_params, skb->protocol);
635 dev->stats.tx_dropped++;
639 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
640 struct net_device *dev)
642 struct ip_tunnel *tunnel = netdev_priv(dev);
644 if (tunnel->collect_md) {
645 gre_fb_xmit(skb, dev);
649 skb = gre_handle_offloads(skb, !!(tunnel->parms.o_flags&TUNNEL_CSUM));
653 if (skb_cow_head(skb, dev->needed_headroom))
656 __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
662 dev->stats.tx_dropped++;
666 static int ipgre_tunnel_ioctl(struct net_device *dev,
667 struct ifreq *ifr, int cmd)
670 struct ip_tunnel_parm p;
672 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
674 if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
675 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
676 p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) ||
677 ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING)))
680 p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
681 p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
683 err = ip_tunnel_ioctl(dev, &p, cmd);
687 p.i_flags = tnl_flags_to_gre_flags(p.i_flags);
688 p.o_flags = tnl_flags_to_gre_flags(p.o_flags);
690 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
695 /* Nice toy. Unfortunately, useless in real life :-)
696 It allows to construct virtual multiprotocol broadcast "LAN"
697 over the Internet, provided multicast routing is tuned.
700 I have no idea was this bicycle invented before me,
701 so that I had to set ARPHRD_IPGRE to a random value.
702 I have an impression, that Cisco could make something similar,
703 but this feature is apparently missing in IOS<=11.2(8).
705 I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
706 with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
708 ping -t 255 224.66.66.66
710 If nobody answers, mbone does not work.
712 ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
713 ip addr add 10.66.66.<somewhat>/24 dev Universe
715 ifconfig Universe add fe80::<Your_real_addr>/10
716 ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
719 ftp fec0:6666:6666::193.233.7.65
722 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
724 const void *daddr, const void *saddr, unsigned int len)
726 struct ip_tunnel *t = netdev_priv(dev);
728 struct gre_base_hdr *greh;
730 iph = (struct iphdr *)skb_push(skb, t->hlen + sizeof(*iph));
731 greh = (struct gre_base_hdr *)(iph+1);
732 greh->flags = tnl_flags_to_gre_flags(t->parms.o_flags);
733 greh->protocol = htons(type);
735 memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
737 /* Set the source hardware address. */
739 memcpy(&iph->saddr, saddr, 4);
741 memcpy(&iph->daddr, daddr, 4);
743 return t->hlen + sizeof(*iph);
745 return -(t->hlen + sizeof(*iph));
748 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
750 const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
751 memcpy(haddr, &iph->saddr, 4);
755 static const struct header_ops ipgre_header_ops = {
756 .create = ipgre_header,
757 .parse = ipgre_header_parse,
760 #ifdef CONFIG_NET_IPGRE_BROADCAST
761 static int ipgre_open(struct net_device *dev)
763 struct ip_tunnel *t = netdev_priv(dev);
765 if (ipv4_is_multicast(t->parms.iph.daddr)) {
769 rt = ip_route_output_gre(t->net, &fl4,
773 RT_TOS(t->parms.iph.tos),
776 return -EADDRNOTAVAIL;
779 if (!__in_dev_get_rtnl(dev))
780 return -EADDRNOTAVAIL;
781 t->mlink = dev->ifindex;
782 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
787 static int ipgre_close(struct net_device *dev)
789 struct ip_tunnel *t = netdev_priv(dev);
791 if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
792 struct in_device *in_dev;
793 in_dev = inetdev_by_index(t->net, t->mlink);
795 ip_mc_dec_group(in_dev, t->parms.iph.daddr);
801 static const struct net_device_ops ipgre_netdev_ops = {
802 .ndo_init = ipgre_tunnel_init,
803 .ndo_uninit = ip_tunnel_uninit,
804 #ifdef CONFIG_NET_IPGRE_BROADCAST
805 .ndo_open = ipgre_open,
806 .ndo_stop = ipgre_close,
808 .ndo_start_xmit = ipgre_xmit,
809 .ndo_do_ioctl = ipgre_tunnel_ioctl,
810 .ndo_change_mtu = ip_tunnel_change_mtu,
811 .ndo_get_stats64 = ip_tunnel_get_stats64,
812 .ndo_get_iflink = ip_tunnel_get_iflink,
815 #define GRE_FEATURES (NETIF_F_SG | \
820 static void ipgre_tunnel_setup(struct net_device *dev)
822 dev->netdev_ops = &ipgre_netdev_ops;
823 dev->type = ARPHRD_IPGRE;
824 ip_tunnel_setup(dev, ipgre_net_id);
827 static void __gre_tunnel_init(struct net_device *dev)
829 struct ip_tunnel *tunnel;
832 tunnel = netdev_priv(dev);
833 tunnel->tun_hlen = ip_gre_calc_hlen(tunnel->parms.o_flags);
834 tunnel->parms.iph.protocol = IPPROTO_GRE;
836 tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
838 t_hlen = tunnel->hlen + sizeof(struct iphdr);
840 dev->needed_headroom = LL_MAX_HEADER + t_hlen + 4;
841 dev->mtu = ETH_DATA_LEN - t_hlen - 4;
843 dev->features |= GRE_FEATURES;
844 dev->hw_features |= GRE_FEATURES;
846 if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
847 /* TCP offload with GRE SEQ is not supported. */
848 dev->features |= NETIF_F_GSO_SOFTWARE;
849 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
850 /* Can use a lockless transmit, unless we generate
853 dev->features |= NETIF_F_LLTX;
857 static int ipgre_tunnel_init(struct net_device *dev)
859 struct ip_tunnel *tunnel = netdev_priv(dev);
860 struct iphdr *iph = &tunnel->parms.iph;
862 __gre_tunnel_init(dev);
864 memcpy(dev->dev_addr, &iph->saddr, 4);
865 memcpy(dev->broadcast, &iph->daddr, 4);
867 dev->flags = IFF_NOARP;
872 #ifdef CONFIG_NET_IPGRE_BROADCAST
873 if (ipv4_is_multicast(iph->daddr)) {
876 dev->flags = IFF_BROADCAST;
877 dev->header_ops = &ipgre_header_ops;
881 dev->header_ops = &ipgre_header_ops;
883 return ip_tunnel_init(dev);
886 static const struct gre_protocol ipgre_protocol = {
888 .err_handler = gre_err,
891 static int __net_init ipgre_init_net(struct net *net)
893 return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
896 static void __net_exit ipgre_exit_net(struct net *net)
898 struct ip_tunnel_net *itn = net_generic(net, ipgre_net_id);
899 ip_tunnel_delete_net(itn, &ipgre_link_ops);
902 static struct pernet_operations ipgre_net_ops = {
903 .init = ipgre_init_net,
904 .exit = ipgre_exit_net,
906 .size = sizeof(struct ip_tunnel_net),
909 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
917 if (data[IFLA_GRE_IFLAGS])
918 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
919 if (data[IFLA_GRE_OFLAGS])
920 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
921 if (flags & (GRE_VERSION|GRE_ROUTING))
927 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
931 if (tb[IFLA_ADDRESS]) {
932 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
934 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
935 return -EADDRNOTAVAIL;
941 if (data[IFLA_GRE_REMOTE]) {
942 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
948 return ipgre_tunnel_validate(tb, data);
951 static void ipgre_netlink_parms(struct net_device *dev,
952 struct nlattr *data[],
954 struct ip_tunnel_parm *parms)
956 memset(parms, 0, sizeof(*parms));
958 parms->iph.protocol = IPPROTO_GRE;
963 if (data[IFLA_GRE_LINK])
964 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
966 if (data[IFLA_GRE_IFLAGS])
967 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
969 if (data[IFLA_GRE_OFLAGS])
970 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
972 if (data[IFLA_GRE_IKEY])
973 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
975 if (data[IFLA_GRE_OKEY])
976 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
978 if (data[IFLA_GRE_LOCAL])
979 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
981 if (data[IFLA_GRE_REMOTE])
982 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
984 if (data[IFLA_GRE_TTL])
985 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
987 if (data[IFLA_GRE_TOS])
988 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
990 if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC]))
991 parms->iph.frag_off = htons(IP_DF);
993 if (data[IFLA_GRE_COLLECT_METADATA]) {
994 struct ip_tunnel *t = netdev_priv(dev);
996 t->collect_md = true;
1000 /* This function returns true when ENCAP attributes are present in the nl msg */
1001 static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1002 struct ip_tunnel_encap *ipencap)
1006 memset(ipencap, 0, sizeof(*ipencap));
1011 if (data[IFLA_GRE_ENCAP_TYPE]) {
1013 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1016 if (data[IFLA_GRE_ENCAP_FLAGS]) {
1018 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1021 if (data[IFLA_GRE_ENCAP_SPORT]) {
1023 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1026 if (data[IFLA_GRE_ENCAP_DPORT]) {
1028 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1034 static int gre_tap_init(struct net_device *dev)
1036 __gre_tunnel_init(dev);
1037 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1039 return ip_tunnel_init(dev);
1042 static const struct net_device_ops gre_tap_netdev_ops = {
1043 .ndo_init = gre_tap_init,
1044 .ndo_uninit = ip_tunnel_uninit,
1045 .ndo_start_xmit = gre_tap_xmit,
1046 .ndo_set_mac_address = eth_mac_addr,
1047 .ndo_validate_addr = eth_validate_addr,
1048 .ndo_change_mtu = ip_tunnel_change_mtu,
1049 .ndo_get_stats64 = ip_tunnel_get_stats64,
1050 .ndo_get_iflink = ip_tunnel_get_iflink,
1051 .ndo_fill_metadata_dst = gre_fill_metadata_dst,
1054 static void ipgre_tap_setup(struct net_device *dev)
1057 dev->netdev_ops = &gre_tap_netdev_ops;
1058 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1059 ip_tunnel_setup(dev, gre_tap_net_id);
1062 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1063 struct nlattr *tb[], struct nlattr *data[])
1065 struct ip_tunnel_parm p;
1066 struct ip_tunnel_encap ipencap;
1068 if (ipgre_netlink_encap_parms(data, &ipencap)) {
1069 struct ip_tunnel *t = netdev_priv(dev);
1070 int err = ip_tunnel_encap_setup(t, &ipencap);
1076 ipgre_netlink_parms(dev, data, tb, &p);
1077 return ip_tunnel_newlink(dev, tb, &p);
1080 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1081 struct nlattr *data[])
1083 struct ip_tunnel_parm p;
1084 struct ip_tunnel_encap ipencap;
1086 if (ipgre_netlink_encap_parms(data, &ipencap)) {
1087 struct ip_tunnel *t = netdev_priv(dev);
1088 int err = ip_tunnel_encap_setup(t, &ipencap);
1094 ipgre_netlink_parms(dev, data, tb, &p);
1095 return ip_tunnel_changelink(dev, tb, &p);
1098 static size_t ipgre_get_size(const struct net_device *dev)
1103 /* IFLA_GRE_IFLAGS */
1105 /* IFLA_GRE_OFLAGS */
1111 /* IFLA_GRE_LOCAL */
1113 /* IFLA_GRE_REMOTE */
1119 /* IFLA_GRE_PMTUDISC */
1121 /* IFLA_GRE_ENCAP_TYPE */
1123 /* IFLA_GRE_ENCAP_FLAGS */
1125 /* IFLA_GRE_ENCAP_SPORT */
1127 /* IFLA_GRE_ENCAP_DPORT */
1129 /* IFLA_GRE_COLLECT_METADATA */
1134 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1136 struct ip_tunnel *t = netdev_priv(dev);
1137 struct ip_tunnel_parm *p = &t->parms;
1139 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1140 nla_put_be16(skb, IFLA_GRE_IFLAGS, tnl_flags_to_gre_flags(p->i_flags)) ||
1141 nla_put_be16(skb, IFLA_GRE_OFLAGS, tnl_flags_to_gre_flags(p->o_flags)) ||
1142 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1143 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1144 nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1145 nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1146 nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1147 nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1148 nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1149 !!(p->iph.frag_off & htons(IP_DF))))
1150 goto nla_put_failure;
1152 if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1154 nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1156 nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1158 nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1160 goto nla_put_failure;
1162 if (t->collect_md) {
1163 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1164 goto nla_put_failure;
1173 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1174 [IFLA_GRE_LINK] = { .type = NLA_U32 },
1175 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 },
1176 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 },
1177 [IFLA_GRE_IKEY] = { .type = NLA_U32 },
1178 [IFLA_GRE_OKEY] = { .type = NLA_U32 },
1179 [IFLA_GRE_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1180 [IFLA_GRE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1181 [IFLA_GRE_TTL] = { .type = NLA_U8 },
1182 [IFLA_GRE_TOS] = { .type = NLA_U8 },
1183 [IFLA_GRE_PMTUDISC] = { .type = NLA_U8 },
1184 [IFLA_GRE_ENCAP_TYPE] = { .type = NLA_U16 },
1185 [IFLA_GRE_ENCAP_FLAGS] = { .type = NLA_U16 },
1186 [IFLA_GRE_ENCAP_SPORT] = { .type = NLA_U16 },
1187 [IFLA_GRE_ENCAP_DPORT] = { .type = NLA_U16 },
1188 [IFLA_GRE_COLLECT_METADATA] = { .type = NLA_FLAG },
1191 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1193 .maxtype = IFLA_GRE_MAX,
1194 .policy = ipgre_policy,
1195 .priv_size = sizeof(struct ip_tunnel),
1196 .setup = ipgre_tunnel_setup,
1197 .validate = ipgre_tunnel_validate,
1198 .newlink = ipgre_newlink,
1199 .changelink = ipgre_changelink,
1200 .dellink = ip_tunnel_dellink,
1201 .get_size = ipgre_get_size,
1202 .fill_info = ipgre_fill_info,
1203 .get_link_net = ip_tunnel_get_link_net,
1206 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1208 .maxtype = IFLA_GRE_MAX,
1209 .policy = ipgre_policy,
1210 .priv_size = sizeof(struct ip_tunnel),
1211 .setup = ipgre_tap_setup,
1212 .validate = ipgre_tap_validate,
1213 .newlink = ipgre_newlink,
1214 .changelink = ipgre_changelink,
1215 .dellink = ip_tunnel_dellink,
1216 .get_size = ipgre_get_size,
1217 .fill_info = ipgre_fill_info,
1218 .get_link_net = ip_tunnel_get_link_net,
1221 struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1222 u8 name_assign_type)
1224 struct nlattr *tb[IFLA_MAX + 1];
1225 struct net_device *dev;
1226 struct ip_tunnel *t;
1229 memset(&tb, 0, sizeof(tb));
1231 dev = rtnl_create_link(net, name, name_assign_type,
1232 &ipgre_tap_ops, tb);
1236 /* Configure flow based GRE device. */
1237 t = netdev_priv(dev);
1238 t->collect_md = true;
1240 err = ipgre_newlink(net, dev, tb, NULL);
1244 /* openvswitch users expect packet sizes to be unrestricted,
1245 * so set the largest MTU we can.
1247 err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1254 return ERR_PTR(err);
1256 EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1258 static int __net_init ipgre_tap_init_net(struct net *net)
1260 return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1263 static void __net_exit ipgre_tap_exit_net(struct net *net)
1265 struct ip_tunnel_net *itn = net_generic(net, gre_tap_net_id);
1266 ip_tunnel_delete_net(itn, &ipgre_tap_ops);
1269 static struct pernet_operations ipgre_tap_net_ops = {
1270 .init = ipgre_tap_init_net,
1271 .exit = ipgre_tap_exit_net,
1272 .id = &gre_tap_net_id,
1273 .size = sizeof(struct ip_tunnel_net),
1276 static int __init ipgre_init(void)
1280 pr_info("GRE over IPv4 tunneling driver\n");
1282 err = register_pernet_device(&ipgre_net_ops);
1286 err = register_pernet_device(&ipgre_tap_net_ops);
1288 goto pnet_tap_faied;
1290 err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1292 pr_info("%s: can't add protocol\n", __func__);
1293 goto add_proto_failed;
1296 err = rtnl_link_register(&ipgre_link_ops);
1298 goto rtnl_link_failed;
1300 err = rtnl_link_register(&ipgre_tap_ops);
1302 goto tap_ops_failed;
1307 rtnl_link_unregister(&ipgre_link_ops);
1309 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1311 unregister_pernet_device(&ipgre_tap_net_ops);
1313 unregister_pernet_device(&ipgre_net_ops);
1317 static void __exit ipgre_fini(void)
1319 rtnl_link_unregister(&ipgre_tap_ops);
1320 rtnl_link_unregister(&ipgre_link_ops);
1321 gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1322 unregister_pernet_device(&ipgre_tap_net_ops);
1323 unregister_pernet_device(&ipgre_net_ops);
1326 module_init(ipgre_init);
1327 module_exit(ipgre_fini);
1328 MODULE_LICENSE("GPL");
1329 MODULE_ALIAS_RTNL_LINK("gre");
1330 MODULE_ALIAS_RTNL_LINK("gretap");
1331 MODULE_ALIAS_NETDEV("gre0");
1332 MODULE_ALIAS_NETDEV("gretap0");