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
, false);
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 __sum16
gre_checksum(struct sk_buff
*skb
)
447 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
448 csum
= lco_csum(skb
);
450 csum
= skb_checksum(skb
, 0, skb
->len
, 0);
451 return csum_fold(csum
);
454 static void build_header(struct sk_buff
*skb
, int hdr_len
, __be16 flags
,
455 __be16 proto
, __be32 key
, __be32 seq
)
457 struct gre_base_hdr
*greh
;
459 skb_push(skb
, hdr_len
);
461 skb_reset_transport_header(skb
);
462 greh
= (struct gre_base_hdr
*)skb
->data
;
463 greh
->flags
= tnl_flags_to_gre_flags(flags
);
464 greh
->protocol
= proto
;
466 if (flags
& (TUNNEL_KEY
| TUNNEL_CSUM
| TUNNEL_SEQ
)) {
467 __be32
*ptr
= (__be32
*)(((u8
*)greh
) + hdr_len
- 4);
469 if (flags
& TUNNEL_SEQ
) {
473 if (flags
& TUNNEL_KEY
) {
477 if (flags
& TUNNEL_CSUM
&&
478 !(skb_shinfo(skb
)->gso_type
&
479 (SKB_GSO_GRE
| SKB_GSO_GRE_CSUM
))) {
481 *(__sum16
*)ptr
= gre_checksum(skb
);
486 static void __gre_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
487 const struct iphdr
*tnl_params
,
490 struct ip_tunnel
*tunnel
= netdev_priv(dev
);
492 if (tunnel
->parms
.o_flags
& TUNNEL_SEQ
)
495 /* Push GRE header. */
496 build_header(skb
, tunnel
->tun_hlen
, tunnel
->parms
.o_flags
,
497 proto
, tunnel
->parms
.o_key
, htonl(tunnel
->o_seqno
));
499 skb_set_inner_protocol(skb
, proto
);
500 ip_tunnel_xmit(skb
, dev
, tnl_params
, tnl_params
->protocol
);
503 static struct sk_buff
*gre_handle_offloads(struct sk_buff
*skb
,
506 return iptunnel_handle_offloads(skb
, csum
? SKB_GSO_GRE_CSUM
: SKB_GSO_GRE
);
509 static struct rtable
*gre_get_rt(struct sk_buff
*skb
,
510 struct net_device
*dev
,
512 const struct ip_tunnel_key
*key
)
514 struct net
*net
= dev_net(dev
);
516 memset(fl
, 0, sizeof(*fl
));
517 fl
->daddr
= key
->u
.ipv4
.dst
;
518 fl
->saddr
= key
->u
.ipv4
.src
;
519 fl
->flowi4_tos
= RT_TOS(key
->tos
);
520 fl
->flowi4_mark
= skb
->mark
;
521 fl
->flowi4_proto
= IPPROTO_GRE
;
523 return ip_route_output_key(net
, fl
);
526 static void gre_fb_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
529 struct ip_tunnel_info
*tun_info
;
530 const struct ip_tunnel_key
*key
;
531 struct rtable
*rt
= NULL
;
539 tun_info
= skb_tunnel_info(skb
);
540 if (unlikely(!tun_info
|| !(tun_info
->mode
& IP_TUNNEL_INFO_TX
) ||
541 ip_tunnel_info_af(tun_info
) != AF_INET
))
544 key
= &tun_info
->key
;
545 use_cache
= ip_tunnel_dst_cache_usable(skb
, tun_info
);
547 rt
= dst_cache_get_ip4(&tun_info
->dst_cache
, &fl
.saddr
);
549 rt
= gre_get_rt(skb
, dev
, &fl
, key
);
553 dst_cache_set_ip4(&tun_info
->dst_cache
, &rt
->dst
,
557 tunnel_hlen
= ip_gre_calc_hlen(key
->tun_flags
);
559 min_headroom
= LL_RESERVED_SPACE(rt
->dst
.dev
) + rt
->dst
.header_len
560 + tunnel_hlen
+ sizeof(struct iphdr
);
561 if (skb_headroom(skb
) < min_headroom
|| skb_header_cloned(skb
)) {
562 int head_delta
= SKB_DATA_ALIGN(min_headroom
-
565 err
= pskb_expand_head(skb
, max_t(int, head_delta
, 0),
571 /* Push Tunnel header. */
572 skb
= gre_handle_offloads(skb
, !!(tun_info
->key
.tun_flags
& TUNNEL_CSUM
));
578 flags
= tun_info
->key
.tun_flags
& (TUNNEL_CSUM
| TUNNEL_KEY
);
579 build_header(skb
, tunnel_hlen
, flags
, proto
,
580 tunnel_id_to_key(tun_info
->key
.tun_id
), 0);
582 df
= key
->tun_flags
& TUNNEL_DONT_FRAGMENT
? htons(IP_DF
) : 0;
584 iptunnel_xmit(skb
->sk
, rt
, skb
, fl
.saddr
, key
->u
.ipv4
.dst
, IPPROTO_GRE
,
585 key
->tos
, key
->ttl
, df
, false);
592 dev
->stats
.tx_dropped
++;
595 static int gre_fill_metadata_dst(struct net_device
*dev
, struct sk_buff
*skb
)
597 struct ip_tunnel_info
*info
= skb_tunnel_info(skb
);
601 if (ip_tunnel_info_af(info
) != AF_INET
)
604 rt
= gre_get_rt(skb
, dev
, &fl4
, &info
->key
);
609 info
->key
.u
.ipv4
.src
= fl4
.saddr
;
613 static netdev_tx_t
ipgre_xmit(struct sk_buff
*skb
,
614 struct net_device
*dev
)
616 struct ip_tunnel
*tunnel
= netdev_priv(dev
);
617 const struct iphdr
*tnl_params
;
619 if (tunnel
->collect_md
) {
620 gre_fb_xmit(skb
, dev
, skb
->protocol
);
624 if (dev
->header_ops
) {
625 /* Need space for new headers */
626 if (skb_cow_head(skb
, dev
->needed_headroom
-
627 (tunnel
->hlen
+ sizeof(struct iphdr
))))
630 tnl_params
= (const struct iphdr
*)skb
->data
;
632 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing
635 skb_pull(skb
, tunnel
->hlen
+ sizeof(struct iphdr
));
636 skb_reset_mac_header(skb
);
638 if (skb_cow_head(skb
, dev
->needed_headroom
))
641 tnl_params
= &tunnel
->parms
.iph
;
644 skb
= gre_handle_offloads(skb
, !!(tunnel
->parms
.o_flags
&TUNNEL_CSUM
));
648 __gre_xmit(skb
, dev
, tnl_params
, skb
->protocol
);
654 dev
->stats
.tx_dropped
++;
658 static netdev_tx_t
gre_tap_xmit(struct sk_buff
*skb
,
659 struct net_device
*dev
)
661 struct ip_tunnel
*tunnel
= netdev_priv(dev
);
663 if (tunnel
->collect_md
) {
664 gre_fb_xmit(skb
, dev
, htons(ETH_P_TEB
));
668 skb
= gre_handle_offloads(skb
, !!(tunnel
->parms
.o_flags
&TUNNEL_CSUM
));
672 if (skb_cow_head(skb
, dev
->needed_headroom
))
675 __gre_xmit(skb
, dev
, &tunnel
->parms
.iph
, htons(ETH_P_TEB
));
681 dev
->stats
.tx_dropped
++;
685 static int ipgre_tunnel_ioctl(struct net_device
*dev
,
686 struct ifreq
*ifr
, int cmd
)
689 struct ip_tunnel_parm p
;
691 if (copy_from_user(&p
, ifr
->ifr_ifru
.ifru_data
, sizeof(p
)))
693 if (cmd
== SIOCADDTUNNEL
|| cmd
== SIOCCHGTUNNEL
) {
694 if (p
.iph
.version
!= 4 || p
.iph
.protocol
!= IPPROTO_GRE
||
695 p
.iph
.ihl
!= 5 || (p
.iph
.frag_off
&htons(~IP_DF
)) ||
696 ((p
.i_flags
|p
.o_flags
)&(GRE_VERSION
|GRE_ROUTING
)))
699 p
.i_flags
= gre_flags_to_tnl_flags(p
.i_flags
);
700 p
.o_flags
= gre_flags_to_tnl_flags(p
.o_flags
);
702 err
= ip_tunnel_ioctl(dev
, &p
, cmd
);
706 p
.i_flags
= tnl_flags_to_gre_flags(p
.i_flags
);
707 p
.o_flags
= tnl_flags_to_gre_flags(p
.o_flags
);
709 if (copy_to_user(ifr
->ifr_ifru
.ifru_data
, &p
, sizeof(p
)))
714 /* Nice toy. Unfortunately, useless in real life :-)
715 It allows to construct virtual multiprotocol broadcast "LAN"
716 over the Internet, provided multicast routing is tuned.
719 I have no idea was this bicycle invented before me,
720 so that I had to set ARPHRD_IPGRE to a random value.
721 I have an impression, that Cisco could make something similar,
722 but this feature is apparently missing in IOS<=11.2(8).
724 I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
725 with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
727 ping -t 255 224.66.66.66
729 If nobody answers, mbone does not work.
731 ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
732 ip addr add 10.66.66.<somewhat>/24 dev Universe
734 ifconfig Universe add fe80::<Your_real_addr>/10
735 ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
738 ftp fec0:6666:6666::193.233.7.65
741 static int ipgre_header(struct sk_buff
*skb
, struct net_device
*dev
,
743 const void *daddr
, const void *saddr
, unsigned int len
)
745 struct ip_tunnel
*t
= netdev_priv(dev
);
747 struct gre_base_hdr
*greh
;
749 iph
= (struct iphdr
*)skb_push(skb
, t
->hlen
+ sizeof(*iph
));
750 greh
= (struct gre_base_hdr
*)(iph
+1);
751 greh
->flags
= tnl_flags_to_gre_flags(t
->parms
.o_flags
);
752 greh
->protocol
= htons(type
);
754 memcpy(iph
, &t
->parms
.iph
, sizeof(struct iphdr
));
756 /* Set the source hardware address. */
758 memcpy(&iph
->saddr
, saddr
, 4);
760 memcpy(&iph
->daddr
, daddr
, 4);
762 return t
->hlen
+ sizeof(*iph
);
764 return -(t
->hlen
+ sizeof(*iph
));
767 static int ipgre_header_parse(const struct sk_buff
*skb
, unsigned char *haddr
)
769 const struct iphdr
*iph
= (const struct iphdr
*) skb_mac_header(skb
);
770 memcpy(haddr
, &iph
->saddr
, 4);
774 static const struct header_ops ipgre_header_ops
= {
775 .create
= ipgre_header
,
776 .parse
= ipgre_header_parse
,
779 #ifdef CONFIG_NET_IPGRE_BROADCAST
780 static int ipgre_open(struct net_device
*dev
)
782 struct ip_tunnel
*t
= netdev_priv(dev
);
784 if (ipv4_is_multicast(t
->parms
.iph
.daddr
)) {
788 rt
= ip_route_output_gre(t
->net
, &fl4
,
792 RT_TOS(t
->parms
.iph
.tos
),
795 return -EADDRNOTAVAIL
;
798 if (!__in_dev_get_rtnl(dev
))
799 return -EADDRNOTAVAIL
;
800 t
->mlink
= dev
->ifindex
;
801 ip_mc_inc_group(__in_dev_get_rtnl(dev
), t
->parms
.iph
.daddr
);
806 static int ipgre_close(struct net_device
*dev
)
808 struct ip_tunnel
*t
= netdev_priv(dev
);
810 if (ipv4_is_multicast(t
->parms
.iph
.daddr
) && t
->mlink
) {
811 struct in_device
*in_dev
;
812 in_dev
= inetdev_by_index(t
->net
, t
->mlink
);
814 ip_mc_dec_group(in_dev
, t
->parms
.iph
.daddr
);
820 static const struct net_device_ops ipgre_netdev_ops
= {
821 .ndo_init
= ipgre_tunnel_init
,
822 .ndo_uninit
= ip_tunnel_uninit
,
823 #ifdef CONFIG_NET_IPGRE_BROADCAST
824 .ndo_open
= ipgre_open
,
825 .ndo_stop
= ipgre_close
,
827 .ndo_start_xmit
= ipgre_xmit
,
828 .ndo_do_ioctl
= ipgre_tunnel_ioctl
,
829 .ndo_change_mtu
= ip_tunnel_change_mtu
,
830 .ndo_get_stats64
= ip_tunnel_get_stats64
,
831 .ndo_get_iflink
= ip_tunnel_get_iflink
,
834 #define GRE_FEATURES (NETIF_F_SG | \
839 static void ipgre_tunnel_setup(struct net_device
*dev
)
841 dev
->netdev_ops
= &ipgre_netdev_ops
;
842 dev
->type
= ARPHRD_IPGRE
;
843 ip_tunnel_setup(dev
, ipgre_net_id
);
846 static void __gre_tunnel_init(struct net_device
*dev
)
848 struct ip_tunnel
*tunnel
;
851 tunnel
= netdev_priv(dev
);
852 tunnel
->tun_hlen
= ip_gre_calc_hlen(tunnel
->parms
.o_flags
);
853 tunnel
->parms
.iph
.protocol
= IPPROTO_GRE
;
855 tunnel
->hlen
= tunnel
->tun_hlen
+ tunnel
->encap_hlen
;
857 t_hlen
= tunnel
->hlen
+ sizeof(struct iphdr
);
859 dev
->needed_headroom
= LL_MAX_HEADER
+ t_hlen
+ 4;
860 dev
->mtu
= ETH_DATA_LEN
- t_hlen
- 4;
862 dev
->features
|= GRE_FEATURES
;
863 dev
->hw_features
|= GRE_FEATURES
;
865 if (!(tunnel
->parms
.o_flags
& TUNNEL_SEQ
)) {
866 /* TCP offload with GRE SEQ is not supported, nor
867 * can we support 2 levels of outer headers requiring
870 if (!(tunnel
->parms
.o_flags
& TUNNEL_CSUM
) ||
871 (tunnel
->encap
.type
== TUNNEL_ENCAP_NONE
)) {
872 dev
->features
|= NETIF_F_GSO_SOFTWARE
;
873 dev
->hw_features
|= NETIF_F_GSO_SOFTWARE
;
876 /* Can use a lockless transmit, unless we generate
879 dev
->features
|= NETIF_F_LLTX
;
883 static int ipgre_tunnel_init(struct net_device
*dev
)
885 struct ip_tunnel
*tunnel
= netdev_priv(dev
);
886 struct iphdr
*iph
= &tunnel
->parms
.iph
;
888 __gre_tunnel_init(dev
);
890 memcpy(dev
->dev_addr
, &iph
->saddr
, 4);
891 memcpy(dev
->broadcast
, &iph
->daddr
, 4);
893 dev
->flags
= IFF_NOARP
;
897 if (iph
->daddr
&& !tunnel
->collect_md
) {
898 #ifdef CONFIG_NET_IPGRE_BROADCAST
899 if (ipv4_is_multicast(iph
->daddr
)) {
902 dev
->flags
= IFF_BROADCAST
;
903 dev
->header_ops
= &ipgre_header_ops
;
906 } else if (!tunnel
->collect_md
) {
907 dev
->header_ops
= &ipgre_header_ops
;
910 return ip_tunnel_init(dev
);
913 static const struct gre_protocol ipgre_protocol
= {
915 .err_handler
= gre_err
,
918 static int __net_init
ipgre_init_net(struct net
*net
)
920 return ip_tunnel_init_net(net
, ipgre_net_id
, &ipgre_link_ops
, NULL
);
923 static void __net_exit
ipgre_exit_net(struct net
*net
)
925 struct ip_tunnel_net
*itn
= net_generic(net
, ipgre_net_id
);
926 ip_tunnel_delete_net(itn
, &ipgre_link_ops
);
929 static struct pernet_operations ipgre_net_ops
= {
930 .init
= ipgre_init_net
,
931 .exit
= ipgre_exit_net
,
933 .size
= sizeof(struct ip_tunnel_net
),
936 static int ipgre_tunnel_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
944 if (data
[IFLA_GRE_IFLAGS
])
945 flags
|= nla_get_be16(data
[IFLA_GRE_IFLAGS
]);
946 if (data
[IFLA_GRE_OFLAGS
])
947 flags
|= nla_get_be16(data
[IFLA_GRE_OFLAGS
]);
948 if (flags
& (GRE_VERSION
|GRE_ROUTING
))
954 static int ipgre_tap_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
958 if (tb
[IFLA_ADDRESS
]) {
959 if (nla_len(tb
[IFLA_ADDRESS
]) != ETH_ALEN
)
961 if (!is_valid_ether_addr(nla_data(tb
[IFLA_ADDRESS
])))
962 return -EADDRNOTAVAIL
;
968 if (data
[IFLA_GRE_REMOTE
]) {
969 memcpy(&daddr
, nla_data(data
[IFLA_GRE_REMOTE
]), 4);
975 return ipgre_tunnel_validate(tb
, data
);
978 static void ipgre_netlink_parms(struct net_device
*dev
,
979 struct nlattr
*data
[],
981 struct ip_tunnel_parm
*parms
)
983 memset(parms
, 0, sizeof(*parms
));
985 parms
->iph
.protocol
= IPPROTO_GRE
;
990 if (data
[IFLA_GRE_LINK
])
991 parms
->link
= nla_get_u32(data
[IFLA_GRE_LINK
]);
993 if (data
[IFLA_GRE_IFLAGS
])
994 parms
->i_flags
= gre_flags_to_tnl_flags(nla_get_be16(data
[IFLA_GRE_IFLAGS
]));
996 if (data
[IFLA_GRE_OFLAGS
])
997 parms
->o_flags
= gre_flags_to_tnl_flags(nla_get_be16(data
[IFLA_GRE_OFLAGS
]));
999 if (data
[IFLA_GRE_IKEY
])
1000 parms
->i_key
= nla_get_be32(data
[IFLA_GRE_IKEY
]);
1002 if (data
[IFLA_GRE_OKEY
])
1003 parms
->o_key
= nla_get_be32(data
[IFLA_GRE_OKEY
]);
1005 if (data
[IFLA_GRE_LOCAL
])
1006 parms
->iph
.saddr
= nla_get_in_addr(data
[IFLA_GRE_LOCAL
]);
1008 if (data
[IFLA_GRE_REMOTE
])
1009 parms
->iph
.daddr
= nla_get_in_addr(data
[IFLA_GRE_REMOTE
]);
1011 if (data
[IFLA_GRE_TTL
])
1012 parms
->iph
.ttl
= nla_get_u8(data
[IFLA_GRE_TTL
]);
1014 if (data
[IFLA_GRE_TOS
])
1015 parms
->iph
.tos
= nla_get_u8(data
[IFLA_GRE_TOS
]);
1017 if (!data
[IFLA_GRE_PMTUDISC
] || nla_get_u8(data
[IFLA_GRE_PMTUDISC
]))
1018 parms
->iph
.frag_off
= htons(IP_DF
);
1020 if (data
[IFLA_GRE_COLLECT_METADATA
]) {
1021 struct ip_tunnel
*t
= netdev_priv(dev
);
1023 t
->collect_md
= true;
1027 /* This function returns true when ENCAP attributes are present in the nl msg */
1028 static bool ipgre_netlink_encap_parms(struct nlattr
*data
[],
1029 struct ip_tunnel_encap
*ipencap
)
1033 memset(ipencap
, 0, sizeof(*ipencap
));
1038 if (data
[IFLA_GRE_ENCAP_TYPE
]) {
1040 ipencap
->type
= nla_get_u16(data
[IFLA_GRE_ENCAP_TYPE
]);
1043 if (data
[IFLA_GRE_ENCAP_FLAGS
]) {
1045 ipencap
->flags
= nla_get_u16(data
[IFLA_GRE_ENCAP_FLAGS
]);
1048 if (data
[IFLA_GRE_ENCAP_SPORT
]) {
1050 ipencap
->sport
= nla_get_be16(data
[IFLA_GRE_ENCAP_SPORT
]);
1053 if (data
[IFLA_GRE_ENCAP_DPORT
]) {
1055 ipencap
->dport
= nla_get_be16(data
[IFLA_GRE_ENCAP_DPORT
]);
1061 static int gre_tap_init(struct net_device
*dev
)
1063 __gre_tunnel_init(dev
);
1064 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
1066 return ip_tunnel_init(dev
);
1069 static const struct net_device_ops gre_tap_netdev_ops
= {
1070 .ndo_init
= gre_tap_init
,
1071 .ndo_uninit
= ip_tunnel_uninit
,
1072 .ndo_start_xmit
= gre_tap_xmit
,
1073 .ndo_set_mac_address
= eth_mac_addr
,
1074 .ndo_validate_addr
= eth_validate_addr
,
1075 .ndo_change_mtu
= ip_tunnel_change_mtu
,
1076 .ndo_get_stats64
= ip_tunnel_get_stats64
,
1077 .ndo_get_iflink
= ip_tunnel_get_iflink
,
1078 .ndo_fill_metadata_dst
= gre_fill_metadata_dst
,
1081 static void ipgre_tap_setup(struct net_device
*dev
)
1084 dev
->netdev_ops
= &gre_tap_netdev_ops
;
1085 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1086 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
1087 ip_tunnel_setup(dev
, gre_tap_net_id
);
1090 static int ipgre_newlink(struct net
*src_net
, struct net_device
*dev
,
1091 struct nlattr
*tb
[], struct nlattr
*data
[])
1093 struct ip_tunnel_parm p
;
1094 struct ip_tunnel_encap ipencap
;
1096 if (ipgre_netlink_encap_parms(data
, &ipencap
)) {
1097 struct ip_tunnel
*t
= netdev_priv(dev
);
1098 int err
= ip_tunnel_encap_setup(t
, &ipencap
);
1104 ipgre_netlink_parms(dev
, data
, tb
, &p
);
1105 return ip_tunnel_newlink(dev
, tb
, &p
);
1108 static int ipgre_changelink(struct net_device
*dev
, struct nlattr
*tb
[],
1109 struct nlattr
*data
[])
1111 struct ip_tunnel_parm p
;
1112 struct ip_tunnel_encap ipencap
;
1114 if (ipgre_netlink_encap_parms(data
, &ipencap
)) {
1115 struct ip_tunnel
*t
= netdev_priv(dev
);
1116 int err
= ip_tunnel_encap_setup(t
, &ipencap
);
1122 ipgre_netlink_parms(dev
, data
, tb
, &p
);
1123 return ip_tunnel_changelink(dev
, tb
, &p
);
1126 static size_t ipgre_get_size(const struct net_device
*dev
)
1131 /* IFLA_GRE_IFLAGS */
1133 /* IFLA_GRE_OFLAGS */
1139 /* IFLA_GRE_LOCAL */
1141 /* IFLA_GRE_REMOTE */
1147 /* IFLA_GRE_PMTUDISC */
1149 /* IFLA_GRE_ENCAP_TYPE */
1151 /* IFLA_GRE_ENCAP_FLAGS */
1153 /* IFLA_GRE_ENCAP_SPORT */
1155 /* IFLA_GRE_ENCAP_DPORT */
1157 /* IFLA_GRE_COLLECT_METADATA */
1162 static int ipgre_fill_info(struct sk_buff
*skb
, const struct net_device
*dev
)
1164 struct ip_tunnel
*t
= netdev_priv(dev
);
1165 struct ip_tunnel_parm
*p
= &t
->parms
;
1167 if (nla_put_u32(skb
, IFLA_GRE_LINK
, p
->link
) ||
1168 nla_put_be16(skb
, IFLA_GRE_IFLAGS
, tnl_flags_to_gre_flags(p
->i_flags
)) ||
1169 nla_put_be16(skb
, IFLA_GRE_OFLAGS
, tnl_flags_to_gre_flags(p
->o_flags
)) ||
1170 nla_put_be32(skb
, IFLA_GRE_IKEY
, p
->i_key
) ||
1171 nla_put_be32(skb
, IFLA_GRE_OKEY
, p
->o_key
) ||
1172 nla_put_in_addr(skb
, IFLA_GRE_LOCAL
, p
->iph
.saddr
) ||
1173 nla_put_in_addr(skb
, IFLA_GRE_REMOTE
, p
->iph
.daddr
) ||
1174 nla_put_u8(skb
, IFLA_GRE_TTL
, p
->iph
.ttl
) ||
1175 nla_put_u8(skb
, IFLA_GRE_TOS
, p
->iph
.tos
) ||
1176 nla_put_u8(skb
, IFLA_GRE_PMTUDISC
,
1177 !!(p
->iph
.frag_off
& htons(IP_DF
))))
1178 goto nla_put_failure
;
1180 if (nla_put_u16(skb
, IFLA_GRE_ENCAP_TYPE
,
1182 nla_put_be16(skb
, IFLA_GRE_ENCAP_SPORT
,
1184 nla_put_be16(skb
, IFLA_GRE_ENCAP_DPORT
,
1186 nla_put_u16(skb
, IFLA_GRE_ENCAP_FLAGS
,
1188 goto nla_put_failure
;
1190 if (t
->collect_md
) {
1191 if (nla_put_flag(skb
, IFLA_GRE_COLLECT_METADATA
))
1192 goto nla_put_failure
;
1201 static const struct nla_policy ipgre_policy
[IFLA_GRE_MAX
+ 1] = {
1202 [IFLA_GRE_LINK
] = { .type
= NLA_U32
},
1203 [IFLA_GRE_IFLAGS
] = { .type
= NLA_U16
},
1204 [IFLA_GRE_OFLAGS
] = { .type
= NLA_U16
},
1205 [IFLA_GRE_IKEY
] = { .type
= NLA_U32
},
1206 [IFLA_GRE_OKEY
] = { .type
= NLA_U32
},
1207 [IFLA_GRE_LOCAL
] = { .len
= FIELD_SIZEOF(struct iphdr
, saddr
) },
1208 [IFLA_GRE_REMOTE
] = { .len
= FIELD_SIZEOF(struct iphdr
, daddr
) },
1209 [IFLA_GRE_TTL
] = { .type
= NLA_U8
},
1210 [IFLA_GRE_TOS
] = { .type
= NLA_U8
},
1211 [IFLA_GRE_PMTUDISC
] = { .type
= NLA_U8
},
1212 [IFLA_GRE_ENCAP_TYPE
] = { .type
= NLA_U16
},
1213 [IFLA_GRE_ENCAP_FLAGS
] = { .type
= NLA_U16
},
1214 [IFLA_GRE_ENCAP_SPORT
] = { .type
= NLA_U16
},
1215 [IFLA_GRE_ENCAP_DPORT
] = { .type
= NLA_U16
},
1216 [IFLA_GRE_COLLECT_METADATA
] = { .type
= NLA_FLAG
},
1219 static struct rtnl_link_ops ipgre_link_ops __read_mostly
= {
1221 .maxtype
= IFLA_GRE_MAX
,
1222 .policy
= ipgre_policy
,
1223 .priv_size
= sizeof(struct ip_tunnel
),
1224 .setup
= ipgre_tunnel_setup
,
1225 .validate
= ipgre_tunnel_validate
,
1226 .newlink
= ipgre_newlink
,
1227 .changelink
= ipgre_changelink
,
1228 .dellink
= ip_tunnel_dellink
,
1229 .get_size
= ipgre_get_size
,
1230 .fill_info
= ipgre_fill_info
,
1231 .get_link_net
= ip_tunnel_get_link_net
,
1234 static struct rtnl_link_ops ipgre_tap_ops __read_mostly
= {
1236 .maxtype
= IFLA_GRE_MAX
,
1237 .policy
= ipgre_policy
,
1238 .priv_size
= sizeof(struct ip_tunnel
),
1239 .setup
= ipgre_tap_setup
,
1240 .validate
= ipgre_tap_validate
,
1241 .newlink
= ipgre_newlink
,
1242 .changelink
= ipgre_changelink
,
1243 .dellink
= ip_tunnel_dellink
,
1244 .get_size
= ipgre_get_size
,
1245 .fill_info
= ipgre_fill_info
,
1246 .get_link_net
= ip_tunnel_get_link_net
,
1249 struct net_device
*gretap_fb_dev_create(struct net
*net
, const char *name
,
1250 u8 name_assign_type
)
1252 struct nlattr
*tb
[IFLA_MAX
+ 1];
1253 struct net_device
*dev
;
1254 struct ip_tunnel
*t
;
1257 memset(&tb
, 0, sizeof(tb
));
1259 dev
= rtnl_create_link(net
, name
, name_assign_type
,
1260 &ipgre_tap_ops
, tb
);
1264 /* Configure flow based GRE device. */
1265 t
= netdev_priv(dev
);
1266 t
->collect_md
= true;
1268 err
= ipgre_newlink(net
, dev
, tb
, NULL
);
1272 /* openvswitch users expect packet sizes to be unrestricted,
1273 * so set the largest MTU we can.
1275 err
= __ip_tunnel_change_mtu(dev
, IP_MAX_MTU
, false);
1282 return ERR_PTR(err
);
1284 EXPORT_SYMBOL_GPL(gretap_fb_dev_create
);
1286 static int __net_init
ipgre_tap_init_net(struct net
*net
)
1288 return ip_tunnel_init_net(net
, gre_tap_net_id
, &ipgre_tap_ops
, "gretap0");
1291 static void __net_exit
ipgre_tap_exit_net(struct net
*net
)
1293 struct ip_tunnel_net
*itn
= net_generic(net
, gre_tap_net_id
);
1294 ip_tunnel_delete_net(itn
, &ipgre_tap_ops
);
1297 static struct pernet_operations ipgre_tap_net_ops
= {
1298 .init
= ipgre_tap_init_net
,
1299 .exit
= ipgre_tap_exit_net
,
1300 .id
= &gre_tap_net_id
,
1301 .size
= sizeof(struct ip_tunnel_net
),
1304 static int __init
ipgre_init(void)
1308 pr_info("GRE over IPv4 tunneling driver\n");
1310 err
= register_pernet_device(&ipgre_net_ops
);
1314 err
= register_pernet_device(&ipgre_tap_net_ops
);
1316 goto pnet_tap_faied
;
1318 err
= gre_add_protocol(&ipgre_protocol
, GREPROTO_CISCO
);
1320 pr_info("%s: can't add protocol\n", __func__
);
1321 goto add_proto_failed
;
1324 err
= rtnl_link_register(&ipgre_link_ops
);
1326 goto rtnl_link_failed
;
1328 err
= rtnl_link_register(&ipgre_tap_ops
);
1330 goto tap_ops_failed
;
1335 rtnl_link_unregister(&ipgre_link_ops
);
1337 gre_del_protocol(&ipgre_protocol
, GREPROTO_CISCO
);
1339 unregister_pernet_device(&ipgre_tap_net_ops
);
1341 unregister_pernet_device(&ipgre_net_ops
);
1345 static void __exit
ipgre_fini(void)
1347 rtnl_link_unregister(&ipgre_tap_ops
);
1348 rtnl_link_unregister(&ipgre_link_ops
);
1349 gre_del_protocol(&ipgre_protocol
, GREPROTO_CISCO
);
1350 unregister_pernet_device(&ipgre_tap_net_ops
);
1351 unregister_pernet_device(&ipgre_net_ops
);
1354 module_init(ipgre_init
);
1355 module_exit(ipgre_fini
);
1356 MODULE_LICENSE("GPL");
1357 MODULE_ALIAS_RTNL_LINK("gre");
1358 MODULE_ALIAS_RTNL_LINK("gretap");
1359 MODULE_ALIAS_NETDEV("gre0");
1360 MODULE_ALIAS_NETDEV("gretap0");