Auto-update from upstream
[linux-2.6] / net / ipv4 / ip_gre.c
1 /*
2  *      Linux NET3:     GRE over IP protocol decoder. 
3  *
4  *      Authors: Alexey Kuznetsov (kuznet@ms2.inr.ac.ru)
5  *
6  *      This program is free software; you can redistribute it and/or
7  *      modify it under the terms of the GNU General Public License
8  *      as published by the Free Software Foundation; either version
9  *      2 of the License, or (at your option) any later version.
10  *
11  */
12
13 #include <linux/config.h>
14 #include <linux/module.h>
15 #include <linux/types.h>
16 #include <linux/sched.h>
17 #include <linux/kernel.h>
18 #include <asm/uaccess.h>
19 #include <linux/skbuff.h>
20 #include <linux/netdevice.h>
21 #include <linux/in.h>
22 #include <linux/tcp.h>
23 #include <linux/udp.h>
24 #include <linux/if_arp.h>
25 #include <linux/mroute.h>
26 #include <linux/init.h>
27 #include <linux/in6.h>
28 #include <linux/inetdevice.h>
29 #include <linux/igmp.h>
30 #include <linux/netfilter_ipv4.h>
31 #include <linux/if_ether.h>
32
33 #include <net/sock.h>
34 #include <net/ip.h>
35 #include <net/icmp.h>
36 #include <net/protocol.h>
37 #include <net/ipip.h>
38 #include <net/arp.h>
39 #include <net/checksum.h>
40 #include <net/dsfield.h>
41 #include <net/inet_ecn.h>
42 #include <net/xfrm.h>
43
44 #ifdef CONFIG_IPV6
45 #include <net/ipv6.h>
46 #include <net/ip6_fib.h>
47 #include <net/ip6_route.h>
48 #endif
49
50 /*
51    Problems & solutions
52    --------------------
53
54    1. The most important issue is detecting local dead loops.
55    They would cause complete host lockup in transmit, which
56    would be "resolved" by stack overflow or, if queueing is enabled,
57    with infinite looping in net_bh.
58
59    We cannot track such dead loops during route installation,
60    it is infeasible task. The most general solutions would be
61    to keep skb->encapsulation counter (sort of local ttl),
62    and silently drop packet when it expires. It is the best
63    solution, but it supposes maintaing new variable in ALL
64    skb, even if no tunneling is used.
65
66    Current solution: t->recursion lock breaks dead loops. It looks 
67    like dev->tbusy flag, but I preferred new variable, because
68    the semantics is different. One day, when hard_start_xmit
69    will be multithreaded we will have to use skb->encapsulation.
70
71
72
73    2. Networking dead loops would not kill routers, but would really
74    kill network. IP hop limit plays role of "t->recursion" in this case,
75    if we copy it from packet being encapsulated to upper header.
76    It is very good solution, but it introduces two problems:
77
78    - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
79      do not work over tunnels.
80    - traceroute does not work. I planned to relay ICMP from tunnel,
81      so that this problem would be solved and traceroute output
82      would even more informative. This idea appeared to be wrong:
83      only Linux complies to rfc1812 now (yes, guys, Linux is the only
84      true router now :-)), all routers (at least, in neighbourhood of mine)
85      return only 8 bytes of payload. It is the end.
86
87    Hence, if we want that OSPF worked or traceroute said something reasonable,
88    we should search for another solution.
89
90    One of them is to parse packet trying to detect inner encapsulation
91    made by our node. It is difficult or even impossible, especially,
92    taking into account fragmentation. TO be short, tt is not solution at all.
93
94    Current solution: The solution was UNEXPECTEDLY SIMPLE.
95    We force DF flag on tunnels with preconfigured hop limit,
96    that is ALL. :-) Well, it does not remove the problem completely,
97    but exponential growth of network traffic is changed to linear
98    (branches, that exceed pmtu are pruned) and tunnel mtu
99    fastly degrades to value <68, where looping stops.
100    Yes, it is not good if there exists a router in the loop,
101    which does not force DF, even when encapsulating packets have DF set.
102    But it is not our problem! Nobody could accuse us, we made
103    all that we could make. Even if it is your gated who injected
104    fatal route to network, even if it were you who configured
105    fatal static route: you are innocent. :-)
106
107
108
109    3. Really, ipv4/ipip.c, ipv4/ip_gre.c and ipv6/sit.c contain
110    practically identical code. It would be good to glue them
111    together, but it is not very evident, how to make them modular.
112    sit is integral part of IPv6, ipip and gre are naturally modular.
113    We could extract common parts (hash table, ioctl etc)
114    to a separate module (ip_tunnel.c).
115
116    Alexey Kuznetsov.
117  */
118
119 static int ipgre_tunnel_init(struct net_device *dev);
120 static void ipgre_tunnel_setup(struct net_device *dev);
121
122 /* Fallback tunnel: no source, no destination, no key, no options */
123
124 static int ipgre_fb_tunnel_init(struct net_device *dev);
125
126 static struct net_device *ipgre_fb_tunnel_dev;
127
128 /* Tunnel hash table */
129
130 /*
131    4 hash tables:
132
133    3: (remote,local)
134    2: (remote,*)
135    1: (*,local)
136    0: (*,*)
137
138    We require exact key match i.e. if a key is present in packet
139    it will match only tunnel with the same key; if it is not present,
140    it will match only keyless tunnel.
141
142    All keysless packets, if not matched configured keyless tunnels
143    will match fallback tunnel.
144  */
145
146 #define HASH_SIZE  16
147 #define HASH(addr) ((addr^(addr>>4))&0xF)
148
149 static struct ip_tunnel *tunnels[4][HASH_SIZE];
150
151 #define tunnels_r_l     (tunnels[3])
152 #define tunnels_r       (tunnels[2])
153 #define tunnels_l       (tunnels[1])
154 #define tunnels_wc      (tunnels[0])
155
156 static DEFINE_RWLOCK(ipgre_lock);
157
158 /* Given src, dst and key, find appropriate for input tunnel. */
159
160 static struct ip_tunnel * ipgre_tunnel_lookup(u32 remote, u32 local, u32 key)
161 {
162         unsigned h0 = HASH(remote);
163         unsigned h1 = HASH(key);
164         struct ip_tunnel *t;
165
166         for (t = tunnels_r_l[h0^h1]; t; t = t->next) {
167                 if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr) {
168                         if (t->parms.i_key == key && (t->dev->flags&IFF_UP))
169                                 return t;
170                 }
171         }
172         for (t = tunnels_r[h0^h1]; t; t = t->next) {
173                 if (remote == t->parms.iph.daddr) {
174                         if (t->parms.i_key == key && (t->dev->flags&IFF_UP))
175                                 return t;
176                 }
177         }
178         for (t = tunnels_l[h1]; t; t = t->next) {
179                 if (local == t->parms.iph.saddr ||
180                      (local == t->parms.iph.daddr && MULTICAST(local))) {
181                         if (t->parms.i_key == key && (t->dev->flags&IFF_UP))
182                                 return t;
183                 }
184         }
185         for (t = tunnels_wc[h1]; t; t = t->next) {
186                 if (t->parms.i_key == key && (t->dev->flags&IFF_UP))
187                         return t;
188         }
189
190         if (ipgre_fb_tunnel_dev->flags&IFF_UP)
191                 return ipgre_fb_tunnel_dev->priv;
192         return NULL;
193 }
194
195 static struct ip_tunnel **ipgre_bucket(struct ip_tunnel *t)
196 {
197         u32 remote = t->parms.iph.daddr;
198         u32 local = t->parms.iph.saddr;
199         u32 key = t->parms.i_key;
200         unsigned h = HASH(key);
201         int prio = 0;
202
203         if (local)
204                 prio |= 1;
205         if (remote && !MULTICAST(remote)) {
206                 prio |= 2;
207                 h ^= HASH(remote);
208         }
209
210         return &tunnels[prio][h];
211 }
212
213 static void ipgre_tunnel_link(struct ip_tunnel *t)
214 {
215         struct ip_tunnel **tp = ipgre_bucket(t);
216
217         t->next = *tp;
218         write_lock_bh(&ipgre_lock);
219         *tp = t;
220         write_unlock_bh(&ipgre_lock);
221 }
222
223 static void ipgre_tunnel_unlink(struct ip_tunnel *t)
224 {
225         struct ip_tunnel **tp;
226
227         for (tp = ipgre_bucket(t); *tp; tp = &(*tp)->next) {
228                 if (t == *tp) {
229                         write_lock_bh(&ipgre_lock);
230                         *tp = t->next;
231                         write_unlock_bh(&ipgre_lock);
232                         break;
233                 }
234         }
235 }
236
237 static struct ip_tunnel * ipgre_tunnel_locate(struct ip_tunnel_parm *parms, int create)
238 {
239         u32 remote = parms->iph.daddr;
240         u32 local = parms->iph.saddr;
241         u32 key = parms->i_key;
242         struct ip_tunnel *t, **tp, *nt;
243         struct net_device *dev;
244         unsigned h = HASH(key);
245         int prio = 0;
246         char name[IFNAMSIZ];
247
248         if (local)
249                 prio |= 1;
250         if (remote && !MULTICAST(remote)) {
251                 prio |= 2;
252                 h ^= HASH(remote);
253         }
254         for (tp = &tunnels[prio][h]; (t = *tp) != NULL; tp = &t->next) {
255                 if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr) {
256                         if (key == t->parms.i_key)
257                                 return t;
258                 }
259         }
260         if (!create)
261                 return NULL;
262
263         if (parms->name[0])
264                 strlcpy(name, parms->name, IFNAMSIZ);
265         else {
266                 int i;
267                 for (i=1; i<100; i++) {
268                         sprintf(name, "gre%d", i);
269                         if (__dev_get_by_name(name) == NULL)
270                                 break;
271                 }
272                 if (i==100)
273                         goto failed;
274         }
275
276         dev = alloc_netdev(sizeof(*t), name, ipgre_tunnel_setup);
277         if (!dev)
278           return NULL;
279
280         dev->init = ipgre_tunnel_init;
281         nt = dev->priv;
282         nt->parms = *parms;
283
284         if (register_netdevice(dev) < 0) {
285                 free_netdev(dev);
286                 goto failed;
287         }
288
289         nt = dev->priv;
290         nt->parms = *parms;
291
292         dev_hold(dev);
293         ipgre_tunnel_link(nt);
294         return nt;
295
296 failed:
297         return NULL;
298 }
299
300 static void ipgre_tunnel_uninit(struct net_device *dev)
301 {
302         ipgre_tunnel_unlink((struct ip_tunnel*)dev->priv);
303         dev_put(dev);
304 }
305
306
307 static void ipgre_err(struct sk_buff *skb, u32 info)
308 {
309 #ifndef I_WISH_WORLD_WERE_PERFECT
310
311 /* It is not :-( All the routers (except for Linux) return only
312    8 bytes of packet payload. It means, that precise relaying of
313    ICMP in the real Internet is absolutely infeasible.
314
315    Moreover, Cisco "wise men" put GRE key to the third word
316    in GRE header. It makes impossible maintaining even soft state for keyed
317    GRE tunnels with enabled checksum. Tell them "thank you".
318
319    Well, I wonder, rfc1812 was written by Cisco employee,
320    what the hell these idiots break standrads established
321    by themself???
322  */
323
324         struct iphdr *iph = (struct iphdr*)skb->data;
325         u16          *p = (u16*)(skb->data+(iph->ihl<<2));
326         int grehlen = (iph->ihl<<2) + 4;
327         int type = skb->h.icmph->type;
328         int code = skb->h.icmph->code;
329         struct ip_tunnel *t;
330         u16 flags;
331
332         flags = p[0];
333         if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
334                 if (flags&(GRE_VERSION|GRE_ROUTING))
335                         return;
336                 if (flags&GRE_KEY) {
337                         grehlen += 4;
338                         if (flags&GRE_CSUM)
339                                 grehlen += 4;
340                 }
341         }
342
343         /* If only 8 bytes returned, keyed message will be dropped here */
344         if (skb_headlen(skb) < grehlen)
345                 return;
346
347         switch (type) {
348         default:
349         case ICMP_PARAMETERPROB:
350                 return;
351
352         case ICMP_DEST_UNREACH:
353                 switch (code) {
354                 case ICMP_SR_FAILED:
355                 case ICMP_PORT_UNREACH:
356                         /* Impossible event. */
357                         return;
358                 case ICMP_FRAG_NEEDED:
359                         /* Soft state for pmtu is maintained by IP core. */
360                         return;
361                 default:
362                         /* All others are translated to HOST_UNREACH.
363                            rfc2003 contains "deep thoughts" about NET_UNREACH,
364                            I believe they are just ether pollution. --ANK
365                          */
366                         break;
367                 }
368                 break;
369         case ICMP_TIME_EXCEEDED:
370                 if (code != ICMP_EXC_TTL)
371                         return;
372                 break;
373         }
374
375         read_lock(&ipgre_lock);
376         t = ipgre_tunnel_lookup(iph->daddr, iph->saddr, (flags&GRE_KEY) ? *(((u32*)p) + (grehlen>>2) - 1) : 0);
377         if (t == NULL || t->parms.iph.daddr == 0 || MULTICAST(t->parms.iph.daddr))
378                 goto out;
379
380         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
381                 goto out;
382
383         if (jiffies - t->err_time < IPTUNNEL_ERR_TIMEO)
384                 t->err_count++;
385         else
386                 t->err_count = 1;
387         t->err_time = jiffies;
388 out:
389         read_unlock(&ipgre_lock);
390         return;
391 #else
392         struct iphdr *iph = (struct iphdr*)dp;
393         struct iphdr *eiph;
394         u16          *p = (u16*)(dp+(iph->ihl<<2));
395         int type = skb->h.icmph->type;
396         int code = skb->h.icmph->code;
397         int rel_type = 0;
398         int rel_code = 0;
399         int rel_info = 0;
400         u16 flags;
401         int grehlen = (iph->ihl<<2) + 4;
402         struct sk_buff *skb2;
403         struct flowi fl;
404         struct rtable *rt;
405
406         if (p[1] != htons(ETH_P_IP))
407                 return;
408
409         flags = p[0];
410         if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
411                 if (flags&(GRE_VERSION|GRE_ROUTING))
412                         return;
413                 if (flags&GRE_CSUM)
414                         grehlen += 4;
415                 if (flags&GRE_KEY)
416                         grehlen += 4;
417                 if (flags&GRE_SEQ)
418                         grehlen += 4;
419         }
420         if (len < grehlen + sizeof(struct iphdr))
421                 return;
422         eiph = (struct iphdr*)(dp + grehlen);
423
424         switch (type) {
425         default:
426                 return;
427         case ICMP_PARAMETERPROB:
428                 if (skb->h.icmph->un.gateway < (iph->ihl<<2))
429                         return;
430
431                 /* So... This guy found something strange INSIDE encapsulated
432                    packet. Well, he is fool, but what can we do ?
433                  */
434                 rel_type = ICMP_PARAMETERPROB;
435                 rel_info = skb->h.icmph->un.gateway - grehlen;
436                 break;
437
438         case ICMP_DEST_UNREACH:
439                 switch (code) {
440                 case ICMP_SR_FAILED:
441                 case ICMP_PORT_UNREACH:
442                         /* Impossible event. */
443                         return;
444                 case ICMP_FRAG_NEEDED:
445                         /* And it is the only really necessary thing :-) */
446                         rel_info = ntohs(skb->h.icmph->un.frag.mtu);
447                         if (rel_info < grehlen+68)
448                                 return;
449                         rel_info -= grehlen;
450                         /* BSD 4.2 MORE DOES NOT EXIST IN NATURE. */
451                         if (rel_info > ntohs(eiph->tot_len))
452                                 return;
453                         break;
454                 default:
455                         /* All others are translated to HOST_UNREACH.
456                            rfc2003 contains "deep thoughts" about NET_UNREACH,
457                            I believe, it is just ether pollution. --ANK
458                          */
459                         rel_type = ICMP_DEST_UNREACH;
460                         rel_code = ICMP_HOST_UNREACH;
461                         break;
462                 }
463                 break;
464         case ICMP_TIME_EXCEEDED:
465                 if (code != ICMP_EXC_TTL)
466                         return;
467                 break;
468         }
469
470         /* Prepare fake skb to feed it to icmp_send */
471         skb2 = skb_clone(skb, GFP_ATOMIC);
472         if (skb2 == NULL)
473                 return;
474         dst_release(skb2->dst);
475         skb2->dst = NULL;
476         skb_pull(skb2, skb->data - (u8*)eiph);
477         skb2->nh.raw = skb2->data;
478
479         /* Try to guess incoming interface */
480         memset(&fl, 0, sizeof(fl));
481         fl.fl4_dst = eiph->saddr;
482         fl.fl4_tos = RT_TOS(eiph->tos);
483         fl.proto = IPPROTO_GRE;
484         if (ip_route_output_key(&rt, &fl)) {
485                 kfree_skb(skb2);
486                 return;
487         }
488         skb2->dev = rt->u.dst.dev;
489
490         /* route "incoming" packet */
491         if (rt->rt_flags&RTCF_LOCAL) {
492                 ip_rt_put(rt);
493                 rt = NULL;
494                 fl.fl4_dst = eiph->daddr;
495                 fl.fl4_src = eiph->saddr;
496                 fl.fl4_tos = eiph->tos;
497                 if (ip_route_output_key(&rt, &fl) ||
498                     rt->u.dst.dev->type != ARPHRD_IPGRE) {
499                         ip_rt_put(rt);
500                         kfree_skb(skb2);
501                         return;
502                 }
503         } else {
504                 ip_rt_put(rt);
505                 if (ip_route_input(skb2, eiph->daddr, eiph->saddr, eiph->tos, skb2->dev) ||
506                     skb2->dst->dev->type != ARPHRD_IPGRE) {
507                         kfree_skb(skb2);
508                         return;
509                 }
510         }
511
512         /* change mtu on this route */
513         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
514                 if (rel_info > dst_mtu(skb2->dst)) {
515                         kfree_skb(skb2);
516                         return;
517                 }
518                 skb2->dst->ops->update_pmtu(skb2->dst, rel_info);
519                 rel_info = htonl(rel_info);
520         } else if (type == ICMP_TIME_EXCEEDED) {
521                 struct ip_tunnel *t = (struct ip_tunnel*)skb2->dev->priv;
522                 if (t->parms.iph.ttl) {
523                         rel_type = ICMP_DEST_UNREACH;
524                         rel_code = ICMP_HOST_UNREACH;
525                 }
526         }
527
528         icmp_send(skb2, rel_type, rel_code, rel_info);
529         kfree_skb(skb2);
530 #endif
531 }
532
533 static inline void ipgre_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb)
534 {
535         if (INET_ECN_is_ce(iph->tos)) {
536                 if (skb->protocol == htons(ETH_P_IP)) {
537                         IP_ECN_set_ce(skb->nh.iph);
538                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
539                         IP6_ECN_set_ce(skb->nh.ipv6h);
540                 }
541         }
542 }
543
544 static inline u8
545 ipgre_ecn_encapsulate(u8 tos, struct iphdr *old_iph, struct sk_buff *skb)
546 {
547         u8 inner = 0;
548         if (skb->protocol == htons(ETH_P_IP))
549                 inner = old_iph->tos;
550         else if (skb->protocol == htons(ETH_P_IPV6))
551                 inner = ipv6_get_dsfield((struct ipv6hdr *)old_iph);
552         return INET_ECN_encapsulate(tos, inner);
553 }
554
555 static int ipgre_rcv(struct sk_buff *skb)
556 {
557         struct iphdr *iph;
558         u8     *h;
559         u16    flags;
560         u16    csum = 0;
561         u32    key = 0;
562         u32    seqno = 0;
563         struct ip_tunnel *tunnel;
564         int    offset = 4;
565
566         if (!pskb_may_pull(skb, 16))
567                 goto drop_nolock;
568
569         iph = skb->nh.iph;
570         h = skb->data;
571         flags = *(u16*)h;
572
573         if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
574                 /* - Version must be 0.
575                    - We do not support routing headers.
576                  */
577                 if (flags&(GRE_VERSION|GRE_ROUTING))
578                         goto drop_nolock;
579
580                 if (flags&GRE_CSUM) {
581                         switch (skb->ip_summed) {
582                         case CHECKSUM_HW:
583                                 csum = (u16)csum_fold(skb->csum);
584                                 if (!csum)
585                                         break;
586                                 /* fall through */
587                         case CHECKSUM_NONE:
588                                 skb->csum = 0;
589                                 csum = __skb_checksum_complete(skb);
590                                 skb->ip_summed = CHECKSUM_HW;
591                         }
592                         offset += 4;
593                 }
594                 if (flags&GRE_KEY) {
595                         key = *(u32*)(h + offset);
596                         offset += 4;
597                 }
598                 if (flags&GRE_SEQ) {
599                         seqno = ntohl(*(u32*)(h + offset));
600                         offset += 4;
601                 }
602         }
603
604         read_lock(&ipgre_lock);
605         if ((tunnel = ipgre_tunnel_lookup(iph->saddr, iph->daddr, key)) != NULL) {
606                 secpath_reset(skb);
607
608                 skb->protocol = *(u16*)(h + 2);
609                 /* WCCP version 1 and 2 protocol decoding.
610                  * - Change protocol to IP
611                  * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
612                  */
613                 if (flags == 0 &&
614                     skb->protocol == __constant_htons(ETH_P_WCCP)) {
615                         skb->protocol = __constant_htons(ETH_P_IP);
616                         if ((*(h + offset) & 0xF0) != 0x40) 
617                                 offset += 4;
618                 }
619
620                 skb->mac.raw = skb->nh.raw;
621                 skb->nh.raw = __pskb_pull(skb, offset);
622                 skb_postpull_rcsum(skb, skb->h.raw, offset);
623                 memset(&(IPCB(skb)->opt), 0, sizeof(struct ip_options));
624                 skb->pkt_type = PACKET_HOST;
625 #ifdef CONFIG_NET_IPGRE_BROADCAST
626                 if (MULTICAST(iph->daddr)) {
627                         /* Looped back packet, drop it! */
628                         if (((struct rtable*)skb->dst)->fl.iif == 0)
629                                 goto drop;
630                         tunnel->stat.multicast++;
631                         skb->pkt_type = PACKET_BROADCAST;
632                 }
633 #endif
634
635                 if (((flags&GRE_CSUM) && csum) ||
636                     (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
637                         tunnel->stat.rx_crc_errors++;
638                         tunnel->stat.rx_errors++;
639                         goto drop;
640                 }
641                 if (tunnel->parms.i_flags&GRE_SEQ) {
642                         if (!(flags&GRE_SEQ) ||
643                             (tunnel->i_seqno && (s32)(seqno - tunnel->i_seqno) < 0)) {
644                                 tunnel->stat.rx_fifo_errors++;
645                                 tunnel->stat.rx_errors++;
646                                 goto drop;
647                         }
648                         tunnel->i_seqno = seqno + 1;
649                 }
650                 tunnel->stat.rx_packets++;
651                 tunnel->stat.rx_bytes += skb->len;
652                 skb->dev = tunnel->dev;
653                 dst_release(skb->dst);
654                 skb->dst = NULL;
655                 nf_reset(skb);
656                 ipgre_ecn_decapsulate(iph, skb);
657                 netif_rx(skb);
658                 read_unlock(&ipgre_lock);
659                 return(0);
660         }
661         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PROT_UNREACH, 0);
662
663 drop:
664         read_unlock(&ipgre_lock);
665 drop_nolock:
666         kfree_skb(skb);
667         return(0);
668 }
669
670 static int ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
671 {
672         struct ip_tunnel *tunnel = (struct ip_tunnel*)dev->priv;
673         struct net_device_stats *stats = &tunnel->stat;
674         struct iphdr  *old_iph = skb->nh.iph;
675         struct iphdr  *tiph;
676         u8     tos;
677         u16    df;
678         struct rtable *rt;                      /* Route to the other host */
679         struct net_device *tdev;                        /* Device to other host */
680         struct iphdr  *iph;                     /* Our new IP header */
681         int    max_headroom;                    /* The extra header space needed */
682         int    gre_hlen;
683         u32    dst;
684         int    mtu;
685
686         if (tunnel->recursion++) {
687                 tunnel->stat.collisions++;
688                 goto tx_error;
689         }
690
691         if (dev->hard_header) {
692                 gre_hlen = 0;
693                 tiph = (struct iphdr*)skb->data;
694         } else {
695                 gre_hlen = tunnel->hlen;
696                 tiph = &tunnel->parms.iph;
697         }
698
699         if ((dst = tiph->daddr) == 0) {
700                 /* NBMA tunnel */
701
702                 if (skb->dst == NULL) {
703                         tunnel->stat.tx_fifo_errors++;
704                         goto tx_error;
705                 }
706
707                 if (skb->protocol == htons(ETH_P_IP)) {
708                         rt = (struct rtable*)skb->dst;
709                         if ((dst = rt->rt_gateway) == 0)
710                                 goto tx_error_icmp;
711                 }
712 #ifdef CONFIG_IPV6
713                 else if (skb->protocol == htons(ETH_P_IPV6)) {
714                         struct in6_addr *addr6;
715                         int addr_type;
716                         struct neighbour *neigh = skb->dst->neighbour;
717
718                         if (neigh == NULL)
719                                 goto tx_error;
720
721                         addr6 = (struct in6_addr*)&neigh->primary_key;
722                         addr_type = ipv6_addr_type(addr6);
723
724                         if (addr_type == IPV6_ADDR_ANY) {
725                                 addr6 = &skb->nh.ipv6h->daddr;
726                                 addr_type = ipv6_addr_type(addr6);
727                         }
728
729                         if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
730                                 goto tx_error_icmp;
731
732                         dst = addr6->s6_addr32[3];
733                 }
734 #endif
735                 else
736                         goto tx_error;
737         }
738
739         tos = tiph->tos;
740         if (tos&1) {
741                 if (skb->protocol == htons(ETH_P_IP))
742                         tos = old_iph->tos;
743                 tos &= ~1;
744         }
745
746         {
747                 struct flowi fl = { .oif = tunnel->parms.link,
748                                     .nl_u = { .ip4_u =
749                                               { .daddr = dst,
750                                                 .saddr = tiph->saddr,
751                                                 .tos = RT_TOS(tos) } },
752                                     .proto = IPPROTO_GRE };
753                 if (ip_route_output_key(&rt, &fl)) {
754                         tunnel->stat.tx_carrier_errors++;
755                         goto tx_error;
756                 }
757         }
758         tdev = rt->u.dst.dev;
759
760         if (tdev == dev) {
761                 ip_rt_put(rt);
762                 tunnel->stat.collisions++;
763                 goto tx_error;
764         }
765
766         df = tiph->frag_off;
767         if (df)
768                 mtu = dst_mtu(&rt->u.dst) - tunnel->hlen;
769         else
770                 mtu = skb->dst ? dst_mtu(skb->dst) : dev->mtu;
771
772         if (skb->dst)
773                 skb->dst->ops->update_pmtu(skb->dst, mtu);
774
775         if (skb->protocol == htons(ETH_P_IP)) {
776                 df |= (old_iph->frag_off&htons(IP_DF));
777
778                 if ((old_iph->frag_off&htons(IP_DF)) &&
779                     mtu < ntohs(old_iph->tot_len)) {
780                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
781                         ip_rt_put(rt);
782                         goto tx_error;
783                 }
784         }
785 #ifdef CONFIG_IPV6
786         else if (skb->protocol == htons(ETH_P_IPV6)) {
787                 struct rt6_info *rt6 = (struct rt6_info*)skb->dst;
788
789                 if (rt6 && mtu < dst_mtu(skb->dst) && mtu >= IPV6_MIN_MTU) {
790                         if ((tunnel->parms.iph.daddr && !MULTICAST(tunnel->parms.iph.daddr)) ||
791                             rt6->rt6i_dst.plen == 128) {
792                                 rt6->rt6i_flags |= RTF_MODIFIED;
793                                 skb->dst->metrics[RTAX_MTU-1] = mtu;
794                         }
795                 }
796
797                 if (mtu >= IPV6_MIN_MTU && mtu < skb->len - tunnel->hlen + gre_hlen) {
798                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, dev);
799                         ip_rt_put(rt);
800                         goto tx_error;
801                 }
802         }
803 #endif
804
805         if (tunnel->err_count > 0) {
806                 if (jiffies - tunnel->err_time < IPTUNNEL_ERR_TIMEO) {
807                         tunnel->err_count--;
808
809                         dst_link_failure(skb);
810                 } else
811                         tunnel->err_count = 0;
812         }
813
814         max_headroom = LL_RESERVED_SPACE(tdev) + gre_hlen;
815
816         if (skb_headroom(skb) < max_headroom || skb_cloned(skb) || skb_shared(skb)) {
817                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
818                 if (!new_skb) {
819                         ip_rt_put(rt);
820                         stats->tx_dropped++;
821                         dev_kfree_skb(skb);
822                         tunnel->recursion--;
823                         return 0;
824                 }
825                 if (skb->sk)
826                         skb_set_owner_w(new_skb, skb->sk);
827                 dev_kfree_skb(skb);
828                 skb = new_skb;
829                 old_iph = skb->nh.iph;
830         }
831
832         skb->h.raw = skb->nh.raw;
833         skb->nh.raw = skb_push(skb, gre_hlen);
834         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
835         dst_release(skb->dst);
836         skb->dst = &rt->u.dst;
837
838         /*
839          *      Push down and install the IPIP header.
840          */
841
842         iph                     =       skb->nh.iph;
843         iph->version            =       4;
844         iph->ihl                =       sizeof(struct iphdr) >> 2;
845         iph->frag_off           =       df;
846         iph->protocol           =       IPPROTO_GRE;
847         iph->tos                =       ipgre_ecn_encapsulate(tos, old_iph, skb);
848         iph->daddr              =       rt->rt_dst;
849         iph->saddr              =       rt->rt_src;
850
851         if ((iph->ttl = tiph->ttl) == 0) {
852                 if (skb->protocol == htons(ETH_P_IP))
853                         iph->ttl = old_iph->ttl;
854 #ifdef CONFIG_IPV6
855                 else if (skb->protocol == htons(ETH_P_IPV6))
856                         iph->ttl = ((struct ipv6hdr*)old_iph)->hop_limit;
857 #endif
858                 else
859                         iph->ttl = dst_metric(&rt->u.dst, RTAX_HOPLIMIT);
860         }
861
862         ((u16*)(iph+1))[0] = tunnel->parms.o_flags;
863         ((u16*)(iph+1))[1] = skb->protocol;
864
865         if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
866                 u32 *ptr = (u32*)(((u8*)iph) + tunnel->hlen - 4);
867
868                 if (tunnel->parms.o_flags&GRE_SEQ) {
869                         ++tunnel->o_seqno;
870                         *ptr = htonl(tunnel->o_seqno);
871                         ptr--;
872                 }
873                 if (tunnel->parms.o_flags&GRE_KEY) {
874                         *ptr = tunnel->parms.o_key;
875                         ptr--;
876                 }
877                 if (tunnel->parms.o_flags&GRE_CSUM) {
878                         *ptr = 0;
879                         *(__u16*)ptr = ip_compute_csum((void*)(iph+1), skb->len - sizeof(struct iphdr));
880                 }
881         }
882
883         nf_reset(skb);
884
885         IPTUNNEL_XMIT();
886         tunnel->recursion--;
887         return 0;
888
889 tx_error_icmp:
890         dst_link_failure(skb);
891
892 tx_error:
893         stats->tx_errors++;
894         dev_kfree_skb(skb);
895         tunnel->recursion--;
896         return 0;
897 }
898
899 static int
900 ipgre_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
901 {
902         int err = 0;
903         struct ip_tunnel_parm p;
904         struct ip_tunnel *t;
905
906         switch (cmd) {
907         case SIOCGETTUNNEL:
908                 t = NULL;
909                 if (dev == ipgre_fb_tunnel_dev) {
910                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
911                                 err = -EFAULT;
912                                 break;
913                         }
914                         t = ipgre_tunnel_locate(&p, 0);
915                 }
916                 if (t == NULL)
917                         t = (struct ip_tunnel*)dev->priv;
918                 memcpy(&p, &t->parms, sizeof(p));
919                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
920                         err = -EFAULT;
921                 break;
922
923         case SIOCADDTUNNEL:
924         case SIOCCHGTUNNEL:
925                 err = -EPERM;
926                 if (!capable(CAP_NET_ADMIN))
927                         goto done;
928
929                 err = -EFAULT;
930                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
931                         goto done;
932
933                 err = -EINVAL;
934                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
935                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) ||
936                     ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING)))
937                         goto done;
938                 if (p.iph.ttl)
939                         p.iph.frag_off |= htons(IP_DF);
940
941                 if (!(p.i_flags&GRE_KEY))
942                         p.i_key = 0;
943                 if (!(p.o_flags&GRE_KEY))
944                         p.o_key = 0;
945
946                 t = ipgre_tunnel_locate(&p, cmd == SIOCADDTUNNEL);
947
948                 if (dev != ipgre_fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
949                         if (t != NULL) {
950                                 if (t->dev != dev) {
951                                         err = -EEXIST;
952                                         break;
953                                 }
954                         } else {
955                                 unsigned nflags=0;
956
957                                 t = (struct ip_tunnel*)dev->priv;
958
959                                 if (MULTICAST(p.iph.daddr))
960                                         nflags = IFF_BROADCAST;
961                                 else if (p.iph.daddr)
962                                         nflags = IFF_POINTOPOINT;
963
964                                 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) {
965                                         err = -EINVAL;
966                                         break;
967                                 }
968                                 ipgre_tunnel_unlink(t);
969                                 t->parms.iph.saddr = p.iph.saddr;
970                                 t->parms.iph.daddr = p.iph.daddr;
971                                 t->parms.i_key = p.i_key;
972                                 t->parms.o_key = p.o_key;
973                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
974                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
975                                 ipgre_tunnel_link(t);
976                                 netdev_state_change(dev);
977                         }
978                 }
979
980                 if (t) {
981                         err = 0;
982                         if (cmd == SIOCCHGTUNNEL) {
983                                 t->parms.iph.ttl = p.iph.ttl;
984                                 t->parms.iph.tos = p.iph.tos;
985                                 t->parms.iph.frag_off = p.iph.frag_off;
986                         }
987                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
988                                 err = -EFAULT;
989                 } else
990                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
991                 break;
992
993         case SIOCDELTUNNEL:
994                 err = -EPERM;
995                 if (!capable(CAP_NET_ADMIN))
996                         goto done;
997
998                 if (dev == ipgre_fb_tunnel_dev) {
999                         err = -EFAULT;
1000                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1001                                 goto done;
1002                         err = -ENOENT;
1003                         if ((t = ipgre_tunnel_locate(&p, 0)) == NULL)
1004                                 goto done;
1005                         err = -EPERM;
1006                         if (t == ipgre_fb_tunnel_dev->priv)
1007                                 goto done;
1008                         dev = t->dev;
1009                 }
1010                 err = unregister_netdevice(dev);
1011                 break;
1012
1013         default:
1014                 err = -EINVAL;
1015         }
1016
1017 done:
1018         return err;
1019 }
1020
1021 static struct net_device_stats *ipgre_tunnel_get_stats(struct net_device *dev)
1022 {
1023         return &(((struct ip_tunnel*)dev->priv)->stat);
1024 }
1025
1026 static int ipgre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1027 {
1028         struct ip_tunnel *tunnel = (struct ip_tunnel*)dev->priv;
1029         if (new_mtu < 68 || new_mtu > 0xFFF8 - tunnel->hlen)
1030                 return -EINVAL;
1031         dev->mtu = new_mtu;
1032         return 0;
1033 }
1034
1035 #ifdef CONFIG_NET_IPGRE_BROADCAST
1036 /* Nice toy. Unfortunately, useless in real life :-)
1037    It allows to construct virtual multiprotocol broadcast "LAN"
1038    over the Internet, provided multicast routing is tuned.
1039
1040
1041    I have no idea was this bicycle invented before me,
1042    so that I had to set ARPHRD_IPGRE to a random value.
1043    I have an impression, that Cisco could make something similar,
1044    but this feature is apparently missing in IOS<=11.2(8).
1045    
1046    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
1047    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
1048
1049    ping -t 255 224.66.66.66
1050
1051    If nobody answers, mbone does not work.
1052
1053    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
1054    ip addr add 10.66.66.<somewhat>/24 dev Universe
1055    ifconfig Universe up
1056    ifconfig Universe add fe80::<Your_real_addr>/10
1057    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
1058    ftp 10.66.66.66
1059    ...
1060    ftp fec0:6666:6666::193.233.7.65
1061    ...
1062
1063  */
1064
1065 static int ipgre_header(struct sk_buff *skb, struct net_device *dev, unsigned short type,
1066                         void *daddr, void *saddr, unsigned len)
1067 {
1068         struct ip_tunnel *t = (struct ip_tunnel*)dev->priv;
1069         struct iphdr *iph = (struct iphdr *)skb_push(skb, t->hlen);
1070         u16 *p = (u16*)(iph+1);
1071
1072         memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
1073         p[0]            = t->parms.o_flags;
1074         p[1]            = htons(type);
1075
1076         /*
1077          *      Set the source hardware address. 
1078          */
1079          
1080         if (saddr)
1081                 memcpy(&iph->saddr, saddr, 4);
1082
1083         if (daddr) {
1084                 memcpy(&iph->daddr, daddr, 4);
1085                 return t->hlen;
1086         }
1087         if (iph->daddr && !MULTICAST(iph->daddr))
1088                 return t->hlen;
1089         
1090         return -t->hlen;
1091 }
1092
1093 static int ipgre_open(struct net_device *dev)
1094 {
1095         struct ip_tunnel *t = (struct ip_tunnel*)dev->priv;
1096
1097         if (MULTICAST(t->parms.iph.daddr)) {
1098                 struct flowi fl = { .oif = t->parms.link,
1099                                     .nl_u = { .ip4_u =
1100                                               { .daddr = t->parms.iph.daddr,
1101                                                 .saddr = t->parms.iph.saddr,
1102                                                 .tos = RT_TOS(t->parms.iph.tos) } },
1103                                     .proto = IPPROTO_GRE };
1104                 struct rtable *rt;
1105                 if (ip_route_output_key(&rt, &fl))
1106                         return -EADDRNOTAVAIL;
1107                 dev = rt->u.dst.dev;
1108                 ip_rt_put(rt);
1109                 if (__in_dev_get_rtnl(dev) == NULL)
1110                         return -EADDRNOTAVAIL;
1111                 t->mlink = dev->ifindex;
1112                 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
1113         }
1114         return 0;
1115 }
1116
1117 static int ipgre_close(struct net_device *dev)
1118 {
1119         struct ip_tunnel *t = (struct ip_tunnel*)dev->priv;
1120         if (MULTICAST(t->parms.iph.daddr) && t->mlink) {
1121                 struct in_device *in_dev = inetdev_by_index(t->mlink);
1122                 if (in_dev) {
1123                         ip_mc_dec_group(in_dev, t->parms.iph.daddr);
1124                         in_dev_put(in_dev);
1125                 }
1126         }
1127         return 0;
1128 }
1129
1130 #endif
1131
1132 static void ipgre_tunnel_setup(struct net_device *dev)
1133 {
1134         SET_MODULE_OWNER(dev);
1135         dev->uninit             = ipgre_tunnel_uninit;
1136         dev->destructor         = free_netdev;
1137         dev->hard_start_xmit    = ipgre_tunnel_xmit;
1138         dev->get_stats          = ipgre_tunnel_get_stats;
1139         dev->do_ioctl           = ipgre_tunnel_ioctl;
1140         dev->change_mtu         = ipgre_tunnel_change_mtu;
1141
1142         dev->type               = ARPHRD_IPGRE;
1143         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr) + 4;
1144         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 4;
1145         dev->flags              = IFF_NOARP;
1146         dev->iflink             = 0;
1147         dev->addr_len           = 4;
1148 }
1149
1150 static int ipgre_tunnel_init(struct net_device *dev)
1151 {
1152         struct net_device *tdev = NULL;
1153         struct ip_tunnel *tunnel;
1154         struct iphdr *iph;
1155         int hlen = LL_MAX_HEADER;
1156         int mtu = ETH_DATA_LEN;
1157         int addend = sizeof(struct iphdr) + 4;
1158
1159         tunnel = (struct ip_tunnel*)dev->priv;
1160         iph = &tunnel->parms.iph;
1161
1162         tunnel->dev = dev;
1163         strcpy(tunnel->parms.name, dev->name);
1164
1165         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1166         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1167
1168         /* Guess output device to choose reasonable mtu and hard_header_len */
1169
1170         if (iph->daddr) {
1171                 struct flowi fl = { .oif = tunnel->parms.link,
1172                                     .nl_u = { .ip4_u =
1173                                               { .daddr = iph->daddr,
1174                                                 .saddr = iph->saddr,
1175                                                 .tos = RT_TOS(iph->tos) } },
1176                                     .proto = IPPROTO_GRE };
1177                 struct rtable *rt;
1178                 if (!ip_route_output_key(&rt, &fl)) {
1179                         tdev = rt->u.dst.dev;
1180                         ip_rt_put(rt);
1181                 }
1182
1183                 dev->flags |= IFF_POINTOPOINT;
1184
1185 #ifdef CONFIG_NET_IPGRE_BROADCAST
1186                 if (MULTICAST(iph->daddr)) {
1187                         if (!iph->saddr)
1188                                 return -EINVAL;
1189                         dev->flags = IFF_BROADCAST;
1190                         dev->hard_header = ipgre_header;
1191                         dev->open = ipgre_open;
1192                         dev->stop = ipgre_close;
1193                 }
1194 #endif
1195         }
1196
1197         if (!tdev && tunnel->parms.link)
1198                 tdev = __dev_get_by_index(tunnel->parms.link);
1199
1200         if (tdev) {
1201                 hlen = tdev->hard_header_len;
1202                 mtu = tdev->mtu;
1203         }
1204         dev->iflink = tunnel->parms.link;
1205
1206         /* Precalculate GRE options length */
1207         if (tunnel->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
1208                 if (tunnel->parms.o_flags&GRE_CSUM)
1209                         addend += 4;
1210                 if (tunnel->parms.o_flags&GRE_KEY)
1211                         addend += 4;
1212                 if (tunnel->parms.o_flags&GRE_SEQ)
1213                         addend += 4;
1214         }
1215         dev->hard_header_len = hlen + addend;
1216         dev->mtu = mtu - addend;
1217         tunnel->hlen = addend;
1218         return 0;
1219 }
1220
1221 static int __init ipgre_fb_tunnel_init(struct net_device *dev)
1222 {
1223         struct ip_tunnel *tunnel = (struct ip_tunnel*)dev->priv;
1224         struct iphdr *iph = &tunnel->parms.iph;
1225
1226         tunnel->dev = dev;
1227         strcpy(tunnel->parms.name, dev->name);
1228
1229         iph->version            = 4;
1230         iph->protocol           = IPPROTO_GRE;
1231         iph->ihl                = 5;
1232         tunnel->hlen            = sizeof(struct iphdr) + 4;
1233
1234         dev_hold(dev);
1235         tunnels_wc[0]           = tunnel;
1236         return 0;
1237 }
1238
1239
1240 static struct net_protocol ipgre_protocol = {
1241         .handler        =       ipgre_rcv,
1242         .err_handler    =       ipgre_err,
1243 };
1244
1245
1246 /*
1247  *      And now the modules code and kernel interface.
1248  */
1249
1250 static int __init ipgre_init(void)
1251 {
1252         int err;
1253
1254         printk(KERN_INFO "GRE over IPv4 tunneling driver\n");
1255
1256         if (inet_add_protocol(&ipgre_protocol, IPPROTO_GRE) < 0) {
1257                 printk(KERN_INFO "ipgre init: can't add protocol\n");
1258                 return -EAGAIN;
1259         }
1260
1261         ipgre_fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "gre0",
1262                                            ipgre_tunnel_setup);
1263         if (!ipgre_fb_tunnel_dev) {
1264                 err = -ENOMEM;
1265                 goto err1;
1266         }
1267
1268         ipgre_fb_tunnel_dev->init = ipgre_fb_tunnel_init;
1269
1270         if ((err = register_netdev(ipgre_fb_tunnel_dev)))
1271                 goto err2;
1272 out:
1273         return err;
1274 err2:
1275         free_netdev(ipgre_fb_tunnel_dev);
1276 err1:
1277         inet_del_protocol(&ipgre_protocol, IPPROTO_GRE);
1278         goto out;
1279 }
1280
1281 static void __exit ipgre_destroy_tunnels(void)
1282 {
1283         int prio;
1284
1285         for (prio = 0; prio < 4; prio++) {
1286                 int h;
1287                 for (h = 0; h < HASH_SIZE; h++) {
1288                         struct ip_tunnel *t;
1289                         while ((t = tunnels[prio][h]) != NULL)
1290                                 unregister_netdevice(t->dev);
1291                 }
1292         }
1293 }
1294
1295 static void __exit ipgre_fini(void)
1296 {
1297         if (inet_del_protocol(&ipgre_protocol, IPPROTO_GRE) < 0)
1298                 printk(KERN_INFO "ipgre close: can't remove protocol\n");
1299
1300         rtnl_lock();
1301         ipgre_destroy_tunnels();
1302         rtnl_unlock();
1303 }
1304
1305 module_init(ipgre_init);
1306 module_exit(ipgre_fini);
1307 MODULE_LICENSE("GPL");