Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[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/capability.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 netdev_priv(ipgre_fb_tunnel_dev);
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 = netdev_priv(dev);
282         nt->parms = *parms;
283
284         if (register_netdevice(dev) < 0) {
285                 free_netdev(dev);
286                 goto failed;
287         }
288
289         dev_hold(dev);
290         ipgre_tunnel_link(nt);
291         return nt;
292
293 failed:
294         return NULL;
295 }
296
297 static void ipgre_tunnel_uninit(struct net_device *dev)
298 {
299         ipgre_tunnel_unlink(netdev_priv(dev));
300         dev_put(dev);
301 }
302
303
304 static void ipgre_err(struct sk_buff *skb, u32 info)
305 {
306 #ifndef I_WISH_WORLD_WERE_PERFECT
307
308 /* It is not :-( All the routers (except for Linux) return only
309    8 bytes of packet payload. It means, that precise relaying of
310    ICMP in the real Internet is absolutely infeasible.
311
312    Moreover, Cisco "wise men" put GRE key to the third word
313    in GRE header. It makes impossible maintaining even soft state for keyed
314    GRE tunnels with enabled checksum. Tell them "thank you".
315
316    Well, I wonder, rfc1812 was written by Cisco employee,
317    what the hell these idiots break standrads established
318    by themself???
319  */
320
321         struct iphdr *iph = (struct iphdr*)skb->data;
322         u16          *p = (u16*)(skb->data+(iph->ihl<<2));
323         int grehlen = (iph->ihl<<2) + 4;
324         int type = skb->h.icmph->type;
325         int code = skb->h.icmph->code;
326         struct ip_tunnel *t;
327         u16 flags;
328
329         flags = p[0];
330         if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
331                 if (flags&(GRE_VERSION|GRE_ROUTING))
332                         return;
333                 if (flags&GRE_KEY) {
334                         grehlen += 4;
335                         if (flags&GRE_CSUM)
336                                 grehlen += 4;
337                 }
338         }
339
340         /* If only 8 bytes returned, keyed message will be dropped here */
341         if (skb_headlen(skb) < grehlen)
342                 return;
343
344         switch (type) {
345         default:
346         case ICMP_PARAMETERPROB:
347                 return;
348
349         case ICMP_DEST_UNREACH:
350                 switch (code) {
351                 case ICMP_SR_FAILED:
352                 case ICMP_PORT_UNREACH:
353                         /* Impossible event. */
354                         return;
355                 case ICMP_FRAG_NEEDED:
356                         /* Soft state for pmtu is maintained by IP core. */
357                         return;
358                 default:
359                         /* All others are translated to HOST_UNREACH.
360                            rfc2003 contains "deep thoughts" about NET_UNREACH,
361                            I believe they are just ether pollution. --ANK
362                          */
363                         break;
364                 }
365                 break;
366         case ICMP_TIME_EXCEEDED:
367                 if (code != ICMP_EXC_TTL)
368                         return;
369                 break;
370         }
371
372         read_lock(&ipgre_lock);
373         t = ipgre_tunnel_lookup(iph->daddr, iph->saddr, (flags&GRE_KEY) ? *(((u32*)p) + (grehlen>>2) - 1) : 0);
374         if (t == NULL || t->parms.iph.daddr == 0 || MULTICAST(t->parms.iph.daddr))
375                 goto out;
376
377         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
378                 goto out;
379
380         if (jiffies - t->err_time < IPTUNNEL_ERR_TIMEO)
381                 t->err_count++;
382         else
383                 t->err_count = 1;
384         t->err_time = jiffies;
385 out:
386         read_unlock(&ipgre_lock);
387         return;
388 #else
389         struct iphdr *iph = (struct iphdr*)dp;
390         struct iphdr *eiph;
391         u16          *p = (u16*)(dp+(iph->ihl<<2));
392         int type = skb->h.icmph->type;
393         int code = skb->h.icmph->code;
394         int rel_type = 0;
395         int rel_code = 0;
396         int rel_info = 0;
397         u16 flags;
398         int grehlen = (iph->ihl<<2) + 4;
399         struct sk_buff *skb2;
400         struct flowi fl;
401         struct rtable *rt;
402
403         if (p[1] != htons(ETH_P_IP))
404                 return;
405
406         flags = p[0];
407         if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
408                 if (flags&(GRE_VERSION|GRE_ROUTING))
409                         return;
410                 if (flags&GRE_CSUM)
411                         grehlen += 4;
412                 if (flags&GRE_KEY)
413                         grehlen += 4;
414                 if (flags&GRE_SEQ)
415                         grehlen += 4;
416         }
417         if (len < grehlen + sizeof(struct iphdr))
418                 return;
419         eiph = (struct iphdr*)(dp + grehlen);
420
421         switch (type) {
422         default:
423                 return;
424         case ICMP_PARAMETERPROB:
425                 if (skb->h.icmph->un.gateway < (iph->ihl<<2))
426                         return;
427
428                 /* So... This guy found something strange INSIDE encapsulated
429                    packet. Well, he is fool, but what can we do ?
430                  */
431                 rel_type = ICMP_PARAMETERPROB;
432                 rel_info = skb->h.icmph->un.gateway - grehlen;
433                 break;
434
435         case ICMP_DEST_UNREACH:
436                 switch (code) {
437                 case ICMP_SR_FAILED:
438                 case ICMP_PORT_UNREACH:
439                         /* Impossible event. */
440                         return;
441                 case ICMP_FRAG_NEEDED:
442                         /* And it is the only really necessary thing :-) */
443                         rel_info = ntohs(skb->h.icmph->un.frag.mtu);
444                         if (rel_info < grehlen+68)
445                                 return;
446                         rel_info -= grehlen;
447                         /* BSD 4.2 MORE DOES NOT EXIST IN NATURE. */
448                         if (rel_info > ntohs(eiph->tot_len))
449                                 return;
450                         break;
451                 default:
452                         /* All others are translated to HOST_UNREACH.
453                            rfc2003 contains "deep thoughts" about NET_UNREACH,
454                            I believe, it is just ether pollution. --ANK
455                          */
456                         rel_type = ICMP_DEST_UNREACH;
457                         rel_code = ICMP_HOST_UNREACH;
458                         break;
459                 }
460                 break;
461         case ICMP_TIME_EXCEEDED:
462                 if (code != ICMP_EXC_TTL)
463                         return;
464                 break;
465         }
466
467         /* Prepare fake skb to feed it to icmp_send */
468         skb2 = skb_clone(skb, GFP_ATOMIC);
469         if (skb2 == NULL)
470                 return;
471         dst_release(skb2->dst);
472         skb2->dst = NULL;
473         skb_pull(skb2, skb->data - (u8*)eiph);
474         skb2->nh.raw = skb2->data;
475
476         /* Try to guess incoming interface */
477         memset(&fl, 0, sizeof(fl));
478         fl.fl4_dst = eiph->saddr;
479         fl.fl4_tos = RT_TOS(eiph->tos);
480         fl.proto = IPPROTO_GRE;
481         if (ip_route_output_key(&rt, &fl)) {
482                 kfree_skb(skb2);
483                 return;
484         }
485         skb2->dev = rt->u.dst.dev;
486
487         /* route "incoming" packet */
488         if (rt->rt_flags&RTCF_LOCAL) {
489                 ip_rt_put(rt);
490                 rt = NULL;
491                 fl.fl4_dst = eiph->daddr;
492                 fl.fl4_src = eiph->saddr;
493                 fl.fl4_tos = eiph->tos;
494                 if (ip_route_output_key(&rt, &fl) ||
495                     rt->u.dst.dev->type != ARPHRD_IPGRE) {
496                         ip_rt_put(rt);
497                         kfree_skb(skb2);
498                         return;
499                 }
500         } else {
501                 ip_rt_put(rt);
502                 if (ip_route_input(skb2, eiph->daddr, eiph->saddr, eiph->tos, skb2->dev) ||
503                     skb2->dst->dev->type != ARPHRD_IPGRE) {
504                         kfree_skb(skb2);
505                         return;
506                 }
507         }
508
509         /* change mtu on this route */
510         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
511                 if (rel_info > dst_mtu(skb2->dst)) {
512                         kfree_skb(skb2);
513                         return;
514                 }
515                 skb2->dst->ops->update_pmtu(skb2->dst, rel_info);
516                 rel_info = htonl(rel_info);
517         } else if (type == ICMP_TIME_EXCEEDED) {
518                 struct ip_tunnel *t = netdev_priv(skb2->dev);
519                 if (t->parms.iph.ttl) {
520                         rel_type = ICMP_DEST_UNREACH;
521                         rel_code = ICMP_HOST_UNREACH;
522                 }
523         }
524
525         icmp_send(skb2, rel_type, rel_code, rel_info);
526         kfree_skb(skb2);
527 #endif
528 }
529
530 static inline void ipgre_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb)
531 {
532         if (INET_ECN_is_ce(iph->tos)) {
533                 if (skb->protocol == htons(ETH_P_IP)) {
534                         IP_ECN_set_ce(skb->nh.iph);
535                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
536                         IP6_ECN_set_ce(skb->nh.ipv6h);
537                 }
538         }
539 }
540
541 static inline u8
542 ipgre_ecn_encapsulate(u8 tos, struct iphdr *old_iph, struct sk_buff *skb)
543 {
544         u8 inner = 0;
545         if (skb->protocol == htons(ETH_P_IP))
546                 inner = old_iph->tos;
547         else if (skb->protocol == htons(ETH_P_IPV6))
548                 inner = ipv6_get_dsfield((struct ipv6hdr *)old_iph);
549         return INET_ECN_encapsulate(tos, inner);
550 }
551
552 static int ipgre_rcv(struct sk_buff *skb)
553 {
554         struct iphdr *iph;
555         u8     *h;
556         u16    flags;
557         u16    csum = 0;
558         u32    key = 0;
559         u32    seqno = 0;
560         struct ip_tunnel *tunnel;
561         int    offset = 4;
562
563         if (!pskb_may_pull(skb, 16))
564                 goto drop_nolock;
565
566         iph = skb->nh.iph;
567         h = skb->data;
568         flags = *(u16*)h;
569
570         if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
571                 /* - Version must be 0.
572                    - We do not support routing headers.
573                  */
574                 if (flags&(GRE_VERSION|GRE_ROUTING))
575                         goto drop_nolock;
576
577                 if (flags&GRE_CSUM) {
578                         switch (skb->ip_summed) {
579                         case CHECKSUM_HW:
580                                 csum = (u16)csum_fold(skb->csum);
581                                 if (!csum)
582                                         break;
583                                 /* fall through */
584                         case CHECKSUM_NONE:
585                                 skb->csum = 0;
586                                 csum = __skb_checksum_complete(skb);
587                                 skb->ip_summed = CHECKSUM_HW;
588                         }
589                         offset += 4;
590                 }
591                 if (flags&GRE_KEY) {
592                         key = *(u32*)(h + offset);
593                         offset += 4;
594                 }
595                 if (flags&GRE_SEQ) {
596                         seqno = ntohl(*(u32*)(h + offset));
597                         offset += 4;
598                 }
599         }
600
601         read_lock(&ipgre_lock);
602         if ((tunnel = ipgre_tunnel_lookup(iph->saddr, iph->daddr, key)) != NULL) {
603                 secpath_reset(skb);
604
605                 skb->protocol = *(u16*)(h + 2);
606                 /* WCCP version 1 and 2 protocol decoding.
607                  * - Change protocol to IP
608                  * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
609                  */
610                 if (flags == 0 &&
611                     skb->protocol == __constant_htons(ETH_P_WCCP)) {
612                         skb->protocol = __constant_htons(ETH_P_IP);
613                         if ((*(h + offset) & 0xF0) != 0x40) 
614                                 offset += 4;
615                 }
616
617                 skb->mac.raw = skb->nh.raw;
618                 skb->nh.raw = __pskb_pull(skb, offset);
619                 skb_postpull_rcsum(skb, skb->h.raw, offset);
620                 memset(&(IPCB(skb)->opt), 0, sizeof(struct ip_options));
621                 skb->pkt_type = PACKET_HOST;
622 #ifdef CONFIG_NET_IPGRE_BROADCAST
623                 if (MULTICAST(iph->daddr)) {
624                         /* Looped back packet, drop it! */
625                         if (((struct rtable*)skb->dst)->fl.iif == 0)
626                                 goto drop;
627                         tunnel->stat.multicast++;
628                         skb->pkt_type = PACKET_BROADCAST;
629                 }
630 #endif
631
632                 if (((flags&GRE_CSUM) && csum) ||
633                     (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
634                         tunnel->stat.rx_crc_errors++;
635                         tunnel->stat.rx_errors++;
636                         goto drop;
637                 }
638                 if (tunnel->parms.i_flags&GRE_SEQ) {
639                         if (!(flags&GRE_SEQ) ||
640                             (tunnel->i_seqno && (s32)(seqno - tunnel->i_seqno) < 0)) {
641                                 tunnel->stat.rx_fifo_errors++;
642                                 tunnel->stat.rx_errors++;
643                                 goto drop;
644                         }
645                         tunnel->i_seqno = seqno + 1;
646                 }
647                 tunnel->stat.rx_packets++;
648                 tunnel->stat.rx_bytes += skb->len;
649                 skb->dev = tunnel->dev;
650                 dst_release(skb->dst);
651                 skb->dst = NULL;
652                 nf_reset(skb);
653                 ipgre_ecn_decapsulate(iph, skb);
654                 netif_rx(skb);
655                 read_unlock(&ipgre_lock);
656                 return(0);
657         }
658         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
659
660 drop:
661         read_unlock(&ipgre_lock);
662 drop_nolock:
663         kfree_skb(skb);
664         return(0);
665 }
666
667 static int ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
668 {
669         struct ip_tunnel *tunnel = netdev_priv(dev);
670         struct net_device_stats *stats = &tunnel->stat;
671         struct iphdr  *old_iph = skb->nh.iph;
672         struct iphdr  *tiph;
673         u8     tos;
674         u16    df;
675         struct rtable *rt;                      /* Route to the other host */
676         struct net_device *tdev;                        /* Device to other host */
677         struct iphdr  *iph;                     /* Our new IP header */
678         int    max_headroom;                    /* The extra header space needed */
679         int    gre_hlen;
680         u32    dst;
681         int    mtu;
682
683         if (tunnel->recursion++) {
684                 tunnel->stat.collisions++;
685                 goto tx_error;
686         }
687
688         if (dev->hard_header) {
689                 gre_hlen = 0;
690                 tiph = (struct iphdr*)skb->data;
691         } else {
692                 gre_hlen = tunnel->hlen;
693                 tiph = &tunnel->parms.iph;
694         }
695
696         if ((dst = tiph->daddr) == 0) {
697                 /* NBMA tunnel */
698
699                 if (skb->dst == NULL) {
700                         tunnel->stat.tx_fifo_errors++;
701                         goto tx_error;
702                 }
703
704                 if (skb->protocol == htons(ETH_P_IP)) {
705                         rt = (struct rtable*)skb->dst;
706                         if ((dst = rt->rt_gateway) == 0)
707                                 goto tx_error_icmp;
708                 }
709 #ifdef CONFIG_IPV6
710                 else if (skb->protocol == htons(ETH_P_IPV6)) {
711                         struct in6_addr *addr6;
712                         int addr_type;
713                         struct neighbour *neigh = skb->dst->neighbour;
714
715                         if (neigh == NULL)
716                                 goto tx_error;
717
718                         addr6 = (struct in6_addr*)&neigh->primary_key;
719                         addr_type = ipv6_addr_type(addr6);
720
721                         if (addr_type == IPV6_ADDR_ANY) {
722                                 addr6 = &skb->nh.ipv6h->daddr;
723                                 addr_type = ipv6_addr_type(addr6);
724                         }
725
726                         if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
727                                 goto tx_error_icmp;
728
729                         dst = addr6->s6_addr32[3];
730                 }
731 #endif
732                 else
733                         goto tx_error;
734         }
735
736         tos = tiph->tos;
737         if (tos&1) {
738                 if (skb->protocol == htons(ETH_P_IP))
739                         tos = old_iph->tos;
740                 tos &= ~1;
741         }
742
743         {
744                 struct flowi fl = { .oif = tunnel->parms.link,
745                                     .nl_u = { .ip4_u =
746                                               { .daddr = dst,
747                                                 .saddr = tiph->saddr,
748                                                 .tos = RT_TOS(tos) } },
749                                     .proto = IPPROTO_GRE };
750                 if (ip_route_output_key(&rt, &fl)) {
751                         tunnel->stat.tx_carrier_errors++;
752                         goto tx_error;
753                 }
754         }
755         tdev = rt->u.dst.dev;
756
757         if (tdev == dev) {
758                 ip_rt_put(rt);
759                 tunnel->stat.collisions++;
760                 goto tx_error;
761         }
762
763         df = tiph->frag_off;
764         if (df)
765                 mtu = dst_mtu(&rt->u.dst) - tunnel->hlen;
766         else
767                 mtu = skb->dst ? dst_mtu(skb->dst) : dev->mtu;
768
769         if (skb->dst)
770                 skb->dst->ops->update_pmtu(skb->dst, mtu);
771
772         if (skb->protocol == htons(ETH_P_IP)) {
773                 df |= (old_iph->frag_off&htons(IP_DF));
774
775                 if ((old_iph->frag_off&htons(IP_DF)) &&
776                     mtu < ntohs(old_iph->tot_len)) {
777                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
778                         ip_rt_put(rt);
779                         goto tx_error;
780                 }
781         }
782 #ifdef CONFIG_IPV6
783         else if (skb->protocol == htons(ETH_P_IPV6)) {
784                 struct rt6_info *rt6 = (struct rt6_info*)skb->dst;
785
786                 if (rt6 && mtu < dst_mtu(skb->dst) && mtu >= IPV6_MIN_MTU) {
787                         if ((tunnel->parms.iph.daddr && !MULTICAST(tunnel->parms.iph.daddr)) ||
788                             rt6->rt6i_dst.plen == 128) {
789                                 rt6->rt6i_flags |= RTF_MODIFIED;
790                                 skb->dst->metrics[RTAX_MTU-1] = mtu;
791                         }
792                 }
793
794                 if (mtu >= IPV6_MIN_MTU && mtu < skb->len - tunnel->hlen + gre_hlen) {
795                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, dev);
796                         ip_rt_put(rt);
797                         goto tx_error;
798                 }
799         }
800 #endif
801
802         if (tunnel->err_count > 0) {
803                 if (jiffies - tunnel->err_time < IPTUNNEL_ERR_TIMEO) {
804                         tunnel->err_count--;
805
806                         dst_link_failure(skb);
807                 } else
808                         tunnel->err_count = 0;
809         }
810
811         max_headroom = LL_RESERVED_SPACE(tdev) + gre_hlen;
812
813         if (skb_headroom(skb) < max_headroom || skb_cloned(skb) || skb_shared(skb)) {
814                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
815                 if (!new_skb) {
816                         ip_rt_put(rt);
817                         stats->tx_dropped++;
818                         dev_kfree_skb(skb);
819                         tunnel->recursion--;
820                         return 0;
821                 }
822                 if (skb->sk)
823                         skb_set_owner_w(new_skb, skb->sk);
824                 dev_kfree_skb(skb);
825                 skb = new_skb;
826                 old_iph = skb->nh.iph;
827         }
828
829         skb->h.raw = skb->nh.raw;
830         skb->nh.raw = skb_push(skb, gre_hlen);
831         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
832         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
833                               IPSKB_REROUTED);
834         dst_release(skb->dst);
835         skb->dst = &rt->u.dst;
836
837         /*
838          *      Push down and install the IPIP header.
839          */
840
841         iph                     =       skb->nh.iph;
842         iph->version            =       4;
843         iph->ihl                =       sizeof(struct iphdr) >> 2;
844         iph->frag_off           =       df;
845         iph->protocol           =       IPPROTO_GRE;
846         iph->tos                =       ipgre_ecn_encapsulate(tos, old_iph, skb);
847         iph->daddr              =       rt->rt_dst;
848         iph->saddr              =       rt->rt_src;
849
850         if ((iph->ttl = tiph->ttl) == 0) {
851                 if (skb->protocol == htons(ETH_P_IP))
852                         iph->ttl = old_iph->ttl;
853 #ifdef CONFIG_IPV6
854                 else if (skb->protocol == htons(ETH_P_IPV6))
855                         iph->ttl = ((struct ipv6hdr*)old_iph)->hop_limit;
856 #endif
857                 else
858                         iph->ttl = dst_metric(&rt->u.dst, RTAX_HOPLIMIT);
859         }
860
861         ((u16*)(iph+1))[0] = tunnel->parms.o_flags;
862         ((u16*)(iph+1))[1] = skb->protocol;
863
864         if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
865                 u32 *ptr = (u32*)(((u8*)iph) + tunnel->hlen - 4);
866
867                 if (tunnel->parms.o_flags&GRE_SEQ) {
868                         ++tunnel->o_seqno;
869                         *ptr = htonl(tunnel->o_seqno);
870                         ptr--;
871                 }
872                 if (tunnel->parms.o_flags&GRE_KEY) {
873                         *ptr = tunnel->parms.o_key;
874                         ptr--;
875                 }
876                 if (tunnel->parms.o_flags&GRE_CSUM) {
877                         *ptr = 0;
878                         *(__u16*)ptr = ip_compute_csum((void*)(iph+1), skb->len - sizeof(struct iphdr));
879                 }
880         }
881
882         nf_reset(skb);
883
884         IPTUNNEL_XMIT();
885         tunnel->recursion--;
886         return 0;
887
888 tx_error_icmp:
889         dst_link_failure(skb);
890
891 tx_error:
892         stats->tx_errors++;
893         dev_kfree_skb(skb);
894         tunnel->recursion--;
895         return 0;
896 }
897
898 static int
899 ipgre_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
900 {
901         int err = 0;
902         struct ip_tunnel_parm p;
903         struct ip_tunnel *t;
904
905         switch (cmd) {
906         case SIOCGETTUNNEL:
907                 t = NULL;
908                 if (dev == ipgre_fb_tunnel_dev) {
909                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
910                                 err = -EFAULT;
911                                 break;
912                         }
913                         t = ipgre_tunnel_locate(&p, 0);
914                 }
915                 if (t == NULL)
916                         t = netdev_priv(dev);
917                 memcpy(&p, &t->parms, sizeof(p));
918                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
919                         err = -EFAULT;
920                 break;
921
922         case SIOCADDTUNNEL:
923         case SIOCCHGTUNNEL:
924                 err = -EPERM;
925                 if (!capable(CAP_NET_ADMIN))
926                         goto done;
927
928                 err = -EFAULT;
929                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
930                         goto done;
931
932                 err = -EINVAL;
933                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
934                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) ||
935                     ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING)))
936                         goto done;
937                 if (p.iph.ttl)
938                         p.iph.frag_off |= htons(IP_DF);
939
940                 if (!(p.i_flags&GRE_KEY))
941                         p.i_key = 0;
942                 if (!(p.o_flags&GRE_KEY))
943                         p.o_key = 0;
944
945                 t = ipgre_tunnel_locate(&p, cmd == SIOCADDTUNNEL);
946
947                 if (dev != ipgre_fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
948                         if (t != NULL) {
949                                 if (t->dev != dev) {
950                                         err = -EEXIST;
951                                         break;
952                                 }
953                         } else {
954                                 unsigned nflags=0;
955
956                                 t = netdev_priv(dev);
957
958                                 if (MULTICAST(p.iph.daddr))
959                                         nflags = IFF_BROADCAST;
960                                 else if (p.iph.daddr)
961                                         nflags = IFF_POINTOPOINT;
962
963                                 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) {
964                                         err = -EINVAL;
965                                         break;
966                                 }
967                                 ipgre_tunnel_unlink(t);
968                                 t->parms.iph.saddr = p.iph.saddr;
969                                 t->parms.iph.daddr = p.iph.daddr;
970                                 t->parms.i_key = p.i_key;
971                                 t->parms.o_key = p.o_key;
972                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
973                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
974                                 ipgre_tunnel_link(t);
975                                 netdev_state_change(dev);
976                         }
977                 }
978
979                 if (t) {
980                         err = 0;
981                         if (cmd == SIOCCHGTUNNEL) {
982                                 t->parms.iph.ttl = p.iph.ttl;
983                                 t->parms.iph.tos = p.iph.tos;
984                                 t->parms.iph.frag_off = p.iph.frag_off;
985                         }
986                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
987                                 err = -EFAULT;
988                 } else
989                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
990                 break;
991
992         case SIOCDELTUNNEL:
993                 err = -EPERM;
994                 if (!capable(CAP_NET_ADMIN))
995                         goto done;
996
997                 if (dev == ipgre_fb_tunnel_dev) {
998                         err = -EFAULT;
999                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1000                                 goto done;
1001                         err = -ENOENT;
1002                         if ((t = ipgre_tunnel_locate(&p, 0)) == NULL)
1003                                 goto done;
1004                         err = -EPERM;
1005                         if (t == netdev_priv(ipgre_fb_tunnel_dev))
1006                                 goto done;
1007                         dev = t->dev;
1008                 }
1009                 err = unregister_netdevice(dev);
1010                 break;
1011
1012         default:
1013                 err = -EINVAL;
1014         }
1015
1016 done:
1017         return err;
1018 }
1019
1020 static struct net_device_stats *ipgre_tunnel_get_stats(struct net_device *dev)
1021 {
1022         return &(((struct ip_tunnel*)netdev_priv(dev))->stat);
1023 }
1024
1025 static int ipgre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1026 {
1027         struct ip_tunnel *tunnel = netdev_priv(dev);
1028         if (new_mtu < 68 || new_mtu > 0xFFF8 - tunnel->hlen)
1029                 return -EINVAL;
1030         dev->mtu = new_mtu;
1031         return 0;
1032 }
1033
1034 #ifdef CONFIG_NET_IPGRE_BROADCAST
1035 /* Nice toy. Unfortunately, useless in real life :-)
1036    It allows to construct virtual multiprotocol broadcast "LAN"
1037    over the Internet, provided multicast routing is tuned.
1038
1039
1040    I have no idea was this bicycle invented before me,
1041    so that I had to set ARPHRD_IPGRE to a random value.
1042    I have an impression, that Cisco could make something similar,
1043    but this feature is apparently missing in IOS<=11.2(8).
1044    
1045    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
1046    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
1047
1048    ping -t 255 224.66.66.66
1049
1050    If nobody answers, mbone does not work.
1051
1052    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
1053    ip addr add 10.66.66.<somewhat>/24 dev Universe
1054    ifconfig Universe up
1055    ifconfig Universe add fe80::<Your_real_addr>/10
1056    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
1057    ftp 10.66.66.66
1058    ...
1059    ftp fec0:6666:6666::193.233.7.65
1060    ...
1061
1062  */
1063
1064 static int ipgre_header(struct sk_buff *skb, struct net_device *dev, unsigned short type,
1065                         void *daddr, void *saddr, unsigned len)
1066 {
1067         struct ip_tunnel *t = netdev_priv(dev);
1068         struct iphdr *iph = (struct iphdr *)skb_push(skb, t->hlen);
1069         u16 *p = (u16*)(iph+1);
1070
1071         memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
1072         p[0]            = t->parms.o_flags;
1073         p[1]            = htons(type);
1074
1075         /*
1076          *      Set the source hardware address. 
1077          */
1078          
1079         if (saddr)
1080                 memcpy(&iph->saddr, saddr, 4);
1081
1082         if (daddr) {
1083                 memcpy(&iph->daddr, daddr, 4);
1084                 return t->hlen;
1085         }
1086         if (iph->daddr && !MULTICAST(iph->daddr))
1087                 return t->hlen;
1088         
1089         return -t->hlen;
1090 }
1091
1092 static int ipgre_open(struct net_device *dev)
1093 {
1094         struct ip_tunnel *t = netdev_priv(dev);
1095
1096         if (MULTICAST(t->parms.iph.daddr)) {
1097                 struct flowi fl = { .oif = t->parms.link,
1098                                     .nl_u = { .ip4_u =
1099                                               { .daddr = t->parms.iph.daddr,
1100                                                 .saddr = t->parms.iph.saddr,
1101                                                 .tos = RT_TOS(t->parms.iph.tos) } },
1102                                     .proto = IPPROTO_GRE };
1103                 struct rtable *rt;
1104                 if (ip_route_output_key(&rt, &fl))
1105                         return -EADDRNOTAVAIL;
1106                 dev = rt->u.dst.dev;
1107                 ip_rt_put(rt);
1108                 if (__in_dev_get_rtnl(dev) == NULL)
1109                         return -EADDRNOTAVAIL;
1110                 t->mlink = dev->ifindex;
1111                 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
1112         }
1113         return 0;
1114 }
1115
1116 static int ipgre_close(struct net_device *dev)
1117 {
1118         struct ip_tunnel *t = netdev_priv(dev);
1119         if (MULTICAST(t->parms.iph.daddr) && t->mlink) {
1120                 struct in_device *in_dev = inetdev_by_index(t->mlink);
1121                 if (in_dev) {
1122                         ip_mc_dec_group(in_dev, t->parms.iph.daddr);
1123                         in_dev_put(in_dev);
1124                 }
1125         }
1126         return 0;
1127 }
1128
1129 #endif
1130
1131 static void ipgre_tunnel_setup(struct net_device *dev)
1132 {
1133         SET_MODULE_OWNER(dev);
1134         dev->uninit             = ipgre_tunnel_uninit;
1135         dev->destructor         = free_netdev;
1136         dev->hard_start_xmit    = ipgre_tunnel_xmit;
1137         dev->get_stats          = ipgre_tunnel_get_stats;
1138         dev->do_ioctl           = ipgre_tunnel_ioctl;
1139         dev->change_mtu         = ipgre_tunnel_change_mtu;
1140
1141         dev->type               = ARPHRD_IPGRE;
1142         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr) + 4;
1143         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 4;
1144         dev->flags              = IFF_NOARP;
1145         dev->iflink             = 0;
1146         dev->addr_len           = 4;
1147 }
1148
1149 static int ipgre_tunnel_init(struct net_device *dev)
1150 {
1151         struct net_device *tdev = NULL;
1152         struct ip_tunnel *tunnel;
1153         struct iphdr *iph;
1154         int hlen = LL_MAX_HEADER;
1155         int mtu = ETH_DATA_LEN;
1156         int addend = sizeof(struct iphdr) + 4;
1157
1158         tunnel = netdev_priv(dev);
1159         iph = &tunnel->parms.iph;
1160
1161         tunnel->dev = dev;
1162         strcpy(tunnel->parms.name, dev->name);
1163
1164         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1165         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1166
1167         /* Guess output device to choose reasonable mtu and hard_header_len */
1168
1169         if (iph->daddr) {
1170                 struct flowi fl = { .oif = tunnel->parms.link,
1171                                     .nl_u = { .ip4_u =
1172                                               { .daddr = iph->daddr,
1173                                                 .saddr = iph->saddr,
1174                                                 .tos = RT_TOS(iph->tos) } },
1175                                     .proto = IPPROTO_GRE };
1176                 struct rtable *rt;
1177                 if (!ip_route_output_key(&rt, &fl)) {
1178                         tdev = rt->u.dst.dev;
1179                         ip_rt_put(rt);
1180                 }
1181
1182                 dev->flags |= IFF_POINTOPOINT;
1183
1184 #ifdef CONFIG_NET_IPGRE_BROADCAST
1185                 if (MULTICAST(iph->daddr)) {
1186                         if (!iph->saddr)
1187                                 return -EINVAL;
1188                         dev->flags = IFF_BROADCAST;
1189                         dev->hard_header = ipgre_header;
1190                         dev->open = ipgre_open;
1191                         dev->stop = ipgre_close;
1192                 }
1193 #endif
1194         }
1195
1196         if (!tdev && tunnel->parms.link)
1197                 tdev = __dev_get_by_index(tunnel->parms.link);
1198
1199         if (tdev) {
1200                 hlen = tdev->hard_header_len;
1201                 mtu = tdev->mtu;
1202         }
1203         dev->iflink = tunnel->parms.link;
1204
1205         /* Precalculate GRE options length */
1206         if (tunnel->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
1207                 if (tunnel->parms.o_flags&GRE_CSUM)
1208                         addend += 4;
1209                 if (tunnel->parms.o_flags&GRE_KEY)
1210                         addend += 4;
1211                 if (tunnel->parms.o_flags&GRE_SEQ)
1212                         addend += 4;
1213         }
1214         dev->hard_header_len = hlen + addend;
1215         dev->mtu = mtu - addend;
1216         tunnel->hlen = addend;
1217         return 0;
1218 }
1219
1220 static int __init ipgre_fb_tunnel_init(struct net_device *dev)
1221 {
1222         struct ip_tunnel *tunnel = netdev_priv(dev);
1223         struct iphdr *iph = &tunnel->parms.iph;
1224
1225         tunnel->dev = dev;
1226         strcpy(tunnel->parms.name, dev->name);
1227
1228         iph->version            = 4;
1229         iph->protocol           = IPPROTO_GRE;
1230         iph->ihl                = 5;
1231         tunnel->hlen            = sizeof(struct iphdr) + 4;
1232
1233         dev_hold(dev);
1234         tunnels_wc[0]           = tunnel;
1235         return 0;
1236 }
1237
1238
1239 static struct net_protocol ipgre_protocol = {
1240         .handler        =       ipgre_rcv,
1241         .err_handler    =       ipgre_err,
1242 };
1243
1244
1245 /*
1246  *      And now the modules code and kernel interface.
1247  */
1248
1249 static int __init ipgre_init(void)
1250 {
1251         int err;
1252
1253         printk(KERN_INFO "GRE over IPv4 tunneling driver\n");
1254
1255         if (inet_add_protocol(&ipgre_protocol, IPPROTO_GRE) < 0) {
1256                 printk(KERN_INFO "ipgre init: can't add protocol\n");
1257                 return -EAGAIN;
1258         }
1259
1260         ipgre_fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "gre0",
1261                                            ipgre_tunnel_setup);
1262         if (!ipgre_fb_tunnel_dev) {
1263                 err = -ENOMEM;
1264                 goto err1;
1265         }
1266
1267         ipgre_fb_tunnel_dev->init = ipgre_fb_tunnel_init;
1268
1269         if ((err = register_netdev(ipgre_fb_tunnel_dev)))
1270                 goto err2;
1271 out:
1272         return err;
1273 err2:
1274         free_netdev(ipgre_fb_tunnel_dev);
1275 err1:
1276         inet_del_protocol(&ipgre_protocol, IPPROTO_GRE);
1277         goto out;
1278 }
1279
1280 static void __exit ipgre_destroy_tunnels(void)
1281 {
1282         int prio;
1283
1284         for (prio = 0; prio < 4; prio++) {
1285                 int h;
1286                 for (h = 0; h < HASH_SIZE; h++) {
1287                         struct ip_tunnel *t;
1288                         while ((t = tunnels[prio][h]) != NULL)
1289                                 unregister_netdevice(t->dev);
1290                 }
1291         }
1292 }
1293
1294 static void __exit ipgre_fini(void)
1295 {
1296         if (inet_del_protocol(&ipgre_protocol, IPPROTO_GRE) < 0)
1297                 printk(KERN_INFO "ipgre close: can't remove protocol\n");
1298
1299         rtnl_lock();
1300         ipgre_destroy_tunnels();
1301         rtnl_unlock();
1302 }
1303
1304 module_init(ipgre_init);
1305 module_exit(ipgre_fini);
1306 MODULE_LICENSE("GPL");