Merge branch 'upstream-davem' of master.kernel.org:/pub/scm/linux/kernel/git/linville...
[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/kernel.h>
17 #include <asm/uaccess.h>
18 #include <linux/skbuff.h>
19 #include <linux/netdevice.h>
20 #include <linux/in.h>
21 #include <linux/tcp.h>
22 #include <linux/udp.h>
23 #include <linux/if_arp.h>
24 #include <linux/mroute.h>
25 #include <linux/init.h>
26 #include <linux/in6.h>
27 #include <linux/inetdevice.h>
28 #include <linux/igmp.h>
29 #include <linux/netfilter_ipv4.h>
30 #include <linux/if_ether.h>
31
32 #include <net/sock.h>
33 #include <net/ip.h>
34 #include <net/icmp.h>
35 #include <net/protocol.h>
36 #include <net/ipip.h>
37 #include <net/arp.h>
38 #include <net/checksum.h>
39 #include <net/dsfield.h>
40 #include <net/inet_ecn.h>
41 #include <net/xfrm.h>
42
43 #ifdef CONFIG_IPV6
44 #include <net/ipv6.h>
45 #include <net/ip6_fib.h>
46 #include <net/ip6_route.h>
47 #endif
48
49 /*
50    Problems & solutions
51    --------------------
52
53    1. The most important issue is detecting local dead loops.
54    They would cause complete host lockup in transmit, which
55    would be "resolved" by stack overflow or, if queueing is enabled,
56    with infinite looping in net_bh.
57
58    We cannot track such dead loops during route installation,
59    it is infeasible task. The most general solutions would be
60    to keep skb->encapsulation counter (sort of local ttl),
61    and silently drop packet when it expires. It is the best
62    solution, but it supposes maintaing new variable in ALL
63    skb, even if no tunneling is used.
64
65    Current solution: t->recursion lock breaks dead loops. It looks
66    like dev->tbusy flag, but I preferred new variable, because
67    the semantics is different. One day, when hard_start_xmit
68    will be multithreaded we will have to use skb->encapsulation.
69
70
71
72    2. Networking dead loops would not kill routers, but would really
73    kill network. IP hop limit plays role of "t->recursion" in this case,
74    if we copy it from packet being encapsulated to upper header.
75    It is very good solution, but it introduces two problems:
76
77    - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
78      do not work over tunnels.
79    - traceroute does not work. I planned to relay ICMP from tunnel,
80      so that this problem would be solved and traceroute output
81      would even more informative. This idea appeared to be wrong:
82      only Linux complies to rfc1812 now (yes, guys, Linux is the only
83      true router now :-)), all routers (at least, in neighbourhood of mine)
84      return only 8 bytes of payload. It is the end.
85
86    Hence, if we want that OSPF worked or traceroute said something reasonable,
87    we should search for another solution.
88
89    One of them is to parse packet trying to detect inner encapsulation
90    made by our node. It is difficult or even impossible, especially,
91    taking into account fragmentation. TO be short, tt is not solution at all.
92
93    Current solution: The solution was UNEXPECTEDLY SIMPLE.
94    We force DF flag on tunnels with preconfigured hop limit,
95    that is ALL. :-) Well, it does not remove the problem completely,
96    but exponential growth of network traffic is changed to linear
97    (branches, that exceed pmtu are pruned) and tunnel mtu
98    fastly degrades to value <68, where looping stops.
99    Yes, it is not good if there exists a router in the loop,
100    which does not force DF, even when encapsulating packets have DF set.
101    But it is not our problem! Nobody could accuse us, we made
102    all that we could make. Even if it is your gated who injected
103    fatal route to network, even if it were you who configured
104    fatal static route: you are innocent. :-)
105
106
107
108    3. Really, ipv4/ipip.c, ipv4/ip_gre.c and ipv6/sit.c contain
109    practically identical code. It would be good to glue them
110    together, but it is not very evident, how to make them modular.
111    sit is integral part of IPv6, ipip and gre are naturally modular.
112    We could extract common parts (hash table, ioctl etc)
113    to a separate module (ip_tunnel.c).
114
115    Alexey Kuznetsov.
116  */
117
118 static int ipgre_tunnel_init(struct net_device *dev);
119 static void ipgre_tunnel_setup(struct net_device *dev);
120
121 /* Fallback tunnel: no source, no destination, no key, no options */
122
123 static int ipgre_fb_tunnel_init(struct net_device *dev);
124
125 static struct net_device *ipgre_fb_tunnel_dev;
126
127 /* Tunnel hash table */
128
129 /*
130    4 hash tables:
131
132    3: (remote,local)
133    2: (remote,*)
134    1: (*,local)
135    0: (*,*)
136
137    We require exact key match i.e. if a key is present in packet
138    it will match only tunnel with the same key; if it is not present,
139    it will match only keyless tunnel.
140
141    All keysless packets, if not matched configured keyless tunnels
142    will match fallback tunnel.
143  */
144
145 #define HASH_SIZE  16
146 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
147
148 static struct ip_tunnel *tunnels[4][HASH_SIZE];
149
150 #define tunnels_r_l     (tunnels[3])
151 #define tunnels_r       (tunnels[2])
152 #define tunnels_l       (tunnels[1])
153 #define tunnels_wc      (tunnels[0])
154
155 static DEFINE_RWLOCK(ipgre_lock);
156
157 /* Given src, dst and key, find appropriate for input tunnel. */
158
159 static struct ip_tunnel * ipgre_tunnel_lookup(__be32 remote, __be32 local, __be32 key)
160 {
161         unsigned h0 = HASH(remote);
162         unsigned h1 = HASH(key);
163         struct ip_tunnel *t;
164
165         for (t = tunnels_r_l[h0^h1]; t; t = t->next) {
166                 if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr) {
167                         if (t->parms.i_key == key && (t->dev->flags&IFF_UP))
168                                 return t;
169                 }
170         }
171         for (t = tunnels_r[h0^h1]; t; t = t->next) {
172                 if (remote == t->parms.iph.daddr) {
173                         if (t->parms.i_key == key && (t->dev->flags&IFF_UP))
174                                 return t;
175                 }
176         }
177         for (t = tunnels_l[h1]; t; t = t->next) {
178                 if (local == t->parms.iph.saddr ||
179                      (local == t->parms.iph.daddr && MULTICAST(local))) {
180                         if (t->parms.i_key == key && (t->dev->flags&IFF_UP))
181                                 return t;
182                 }
183         }
184         for (t = tunnels_wc[h1]; t; t = t->next) {
185                 if (t->parms.i_key == key && (t->dev->flags&IFF_UP))
186                         return t;
187         }
188
189         if (ipgre_fb_tunnel_dev->flags&IFF_UP)
190                 return netdev_priv(ipgre_fb_tunnel_dev);
191         return NULL;
192 }
193
194 static struct ip_tunnel **__ipgre_bucket(struct ip_tunnel_parm *parms)
195 {
196         __be32 remote = parms->iph.daddr;
197         __be32 local = parms->iph.saddr;
198         __be32 key = parms->i_key;
199         unsigned h = HASH(key);
200         int prio = 0;
201
202         if (local)
203                 prio |= 1;
204         if (remote && !MULTICAST(remote)) {
205                 prio |= 2;
206                 h ^= HASH(remote);
207         }
208
209         return &tunnels[prio][h];
210 }
211
212 static inline struct ip_tunnel **ipgre_bucket(struct ip_tunnel *t)
213 {
214         return __ipgre_bucket(&t->parms);
215 }
216
217 static void ipgre_tunnel_link(struct ip_tunnel *t)
218 {
219         struct ip_tunnel **tp = ipgre_bucket(t);
220
221         t->next = *tp;
222         write_lock_bh(&ipgre_lock);
223         *tp = t;
224         write_unlock_bh(&ipgre_lock);
225 }
226
227 static void ipgre_tunnel_unlink(struct ip_tunnel *t)
228 {
229         struct ip_tunnel **tp;
230
231         for (tp = ipgre_bucket(t); *tp; tp = &(*tp)->next) {
232                 if (t == *tp) {
233                         write_lock_bh(&ipgre_lock);
234                         *tp = t->next;
235                         write_unlock_bh(&ipgre_lock);
236                         break;
237                 }
238         }
239 }
240
241 static struct ip_tunnel * ipgre_tunnel_locate(struct ip_tunnel_parm *parms, int create)
242 {
243         __be32 remote = parms->iph.daddr;
244         __be32 local = parms->iph.saddr;
245         __be32 key = parms->i_key;
246         struct ip_tunnel *t, **tp, *nt;
247         struct net_device *dev;
248         char name[IFNAMSIZ];
249
250         for (tp = __ipgre_bucket(parms); (t = *tp) != NULL; tp = &t->next) {
251                 if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr) {
252                         if (key == t->parms.i_key)
253                                 return t;
254                 }
255         }
256         if (!create)
257                 return NULL;
258
259         if (parms->name[0])
260                 strlcpy(name, parms->name, IFNAMSIZ);
261         else {
262                 int i;
263                 for (i=1; i<100; i++) {
264                         sprintf(name, "gre%d", i);
265                         if (__dev_get_by_name(name) == NULL)
266                                 break;
267                 }
268                 if (i==100)
269                         goto failed;
270         }
271
272         dev = alloc_netdev(sizeof(*t), name, ipgre_tunnel_setup);
273         if (!dev)
274           return NULL;
275
276         dev->init = ipgre_tunnel_init;
277         nt = netdev_priv(dev);
278         nt->parms = *parms;
279
280         if (register_netdevice(dev) < 0) {
281                 free_netdev(dev);
282                 goto failed;
283         }
284
285         dev_hold(dev);
286         ipgre_tunnel_link(nt);
287         return nt;
288
289 failed:
290         return NULL;
291 }
292
293 static void ipgre_tunnel_uninit(struct net_device *dev)
294 {
295         ipgre_tunnel_unlink(netdev_priv(dev));
296         dev_put(dev);
297 }
298
299
300 static void ipgre_err(struct sk_buff *skb, u32 info)
301 {
302 #ifndef I_WISH_WORLD_WERE_PERFECT
303
304 /* It is not :-( All the routers (except for Linux) return only
305    8 bytes of packet payload. It means, that precise relaying of
306    ICMP in the real Internet is absolutely infeasible.
307
308    Moreover, Cisco "wise men" put GRE key to the third word
309    in GRE header. It makes impossible maintaining even soft state for keyed
310    GRE tunnels with enabled checksum. Tell them "thank you".
311
312    Well, I wonder, rfc1812 was written by Cisco employee,
313    what the hell these idiots break standrads established
314    by themself???
315  */
316
317         struct iphdr *iph = (struct iphdr*)skb->data;
318         __be16       *p = (__be16*)(skb->data+(iph->ihl<<2));
319         int grehlen = (iph->ihl<<2) + 4;
320         const int type = icmp_hdr(skb)->type;
321         const int code = icmp_hdr(skb)->code;
322         struct ip_tunnel *t;
323         __be16 flags;
324
325         flags = p[0];
326         if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
327                 if (flags&(GRE_VERSION|GRE_ROUTING))
328                         return;
329                 if (flags&GRE_KEY) {
330                         grehlen += 4;
331                         if (flags&GRE_CSUM)
332                                 grehlen += 4;
333                 }
334         }
335
336         /* If only 8 bytes returned, keyed message will be dropped here */
337         if (skb_headlen(skb) < grehlen)
338                 return;
339
340         switch (type) {
341         default:
342         case ICMP_PARAMETERPROB:
343                 return;
344
345         case ICMP_DEST_UNREACH:
346                 switch (code) {
347                 case ICMP_SR_FAILED:
348                 case ICMP_PORT_UNREACH:
349                         /* Impossible event. */
350                         return;
351                 case ICMP_FRAG_NEEDED:
352                         /* Soft state for pmtu is maintained by IP core. */
353                         return;
354                 default:
355                         /* All others are translated to HOST_UNREACH.
356                            rfc2003 contains "deep thoughts" about NET_UNREACH,
357                            I believe they are just ether pollution. --ANK
358                          */
359                         break;
360                 }
361                 break;
362         case ICMP_TIME_EXCEEDED:
363                 if (code != ICMP_EXC_TTL)
364                         return;
365                 break;
366         }
367
368         read_lock(&ipgre_lock);
369         t = ipgre_tunnel_lookup(iph->daddr, iph->saddr, (flags&GRE_KEY) ? *(((__be32*)p) + (grehlen>>2) - 1) : 0);
370         if (t == NULL || t->parms.iph.daddr == 0 || MULTICAST(t->parms.iph.daddr))
371                 goto out;
372
373         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
374                 goto out;
375
376         if (jiffies - t->err_time < IPTUNNEL_ERR_TIMEO)
377                 t->err_count++;
378         else
379                 t->err_count = 1;
380         t->err_time = jiffies;
381 out:
382         read_unlock(&ipgre_lock);
383         return;
384 #else
385         struct iphdr *iph = (struct iphdr*)dp;
386         struct iphdr *eiph;
387         __be16       *p = (__be16*)(dp+(iph->ihl<<2));
388         const int type = icmp_hdr(skb)->type;
389         const int code = icmp_hdr(skb)->code;
390         int rel_type = 0;
391         int rel_code = 0;
392         __be32 rel_info = 0;
393         __u32 n = 0;
394         __be16 flags;
395         int grehlen = (iph->ihl<<2) + 4;
396         struct sk_buff *skb2;
397         struct flowi fl;
398         struct rtable *rt;
399
400         if (p[1] != htons(ETH_P_IP))
401                 return;
402
403         flags = p[0];
404         if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
405                 if (flags&(GRE_VERSION|GRE_ROUTING))
406                         return;
407                 if (flags&GRE_CSUM)
408                         grehlen += 4;
409                 if (flags&GRE_KEY)
410                         grehlen += 4;
411                 if (flags&GRE_SEQ)
412                         grehlen += 4;
413         }
414         if (len < grehlen + sizeof(struct iphdr))
415                 return;
416         eiph = (struct iphdr*)(dp + grehlen);
417
418         switch (type) {
419         default:
420                 return;
421         case ICMP_PARAMETERPROB:
422                 n = ntohl(icmp_hdr(skb)->un.gateway) >> 24;
423                 if (n < (iph->ihl<<2))
424                         return;
425
426                 /* So... This guy found something strange INSIDE encapsulated
427                    packet. Well, he is fool, but what can we do ?
428                  */
429                 rel_type = ICMP_PARAMETERPROB;
430                 n -= grehlen;
431                 rel_info = htonl(n << 24);
432                 break;
433
434         case ICMP_DEST_UNREACH:
435                 switch (code) {
436                 case ICMP_SR_FAILED:
437                 case ICMP_PORT_UNREACH:
438                         /* Impossible event. */
439                         return;
440                 case ICMP_FRAG_NEEDED:
441                         /* And it is the only really necessary thing :-) */
442                         n = ntohs(icmp_hdr(skb)->un.frag.mtu);
443                         if (n < grehlen+68)
444                                 return;
445                         n -= grehlen;
446                         /* BSD 4.2 MORE DOES NOT EXIST IN NATURE. */
447                         if (n > ntohs(eiph->tot_len))
448                                 return;
449                         rel_info = htonl(n);
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         skb_reset_network_header(skb2);
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 (n > dst_mtu(skb2->dst)) {
512                         kfree_skb(skb2);
513                         return;
514                 }
515                 skb2->dst->ops->update_pmtu(skb2->dst, n);
516         } else if (type == ICMP_TIME_EXCEEDED) {
517                 struct ip_tunnel *t = netdev_priv(skb2->dev);
518                 if (t->parms.iph.ttl) {
519                         rel_type = ICMP_DEST_UNREACH;
520                         rel_code = ICMP_HOST_UNREACH;
521                 }
522         }
523
524         icmp_send(skb2, rel_type, rel_code, rel_info);
525         kfree_skb(skb2);
526 #endif
527 }
528
529 static inline void ipgre_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb)
530 {
531         if (INET_ECN_is_ce(iph->tos)) {
532                 if (skb->protocol == htons(ETH_P_IP)) {
533                         IP_ECN_set_ce(ip_hdr(skb));
534                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
535                         IP6_ECN_set_ce(ipv6_hdr(skb));
536                 }
537         }
538 }
539
540 static inline u8
541 ipgre_ecn_encapsulate(u8 tos, struct iphdr *old_iph, struct sk_buff *skb)
542 {
543         u8 inner = 0;
544         if (skb->protocol == htons(ETH_P_IP))
545                 inner = old_iph->tos;
546         else if (skb->protocol == htons(ETH_P_IPV6))
547                 inner = ipv6_get_dsfield((struct ipv6hdr *)old_iph);
548         return INET_ECN_encapsulate(tos, inner);
549 }
550
551 static int ipgre_rcv(struct sk_buff *skb)
552 {
553         struct iphdr *iph;
554         u8     *h;
555         __be16    flags;
556         __sum16   csum = 0;
557         __be32 key = 0;
558         u32    seqno = 0;
559         struct ip_tunnel *tunnel;
560         int    offset = 4;
561
562         if (!pskb_may_pull(skb, 16))
563                 goto drop_nolock;
564
565         iph = ip_hdr(skb);
566         h = skb->data;
567         flags = *(__be16*)h;
568
569         if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
570                 /* - Version must be 0.
571                    - We do not support routing headers.
572                  */
573                 if (flags&(GRE_VERSION|GRE_ROUTING))
574                         goto drop_nolock;
575
576                 if (flags&GRE_CSUM) {
577                         switch (skb->ip_summed) {
578                         case CHECKSUM_COMPLETE:
579                                 csum = csum_fold(skb->csum);
580                                 if (!csum)
581                                         break;
582                                 /* fall through */
583                         case CHECKSUM_NONE:
584                                 skb->csum = 0;
585                                 csum = __skb_checksum_complete(skb);
586                                 skb->ip_summed = CHECKSUM_COMPLETE;
587                         }
588                         offset += 4;
589                 }
590                 if (flags&GRE_KEY) {
591                         key = *(__be32*)(h + offset);
592                         offset += 4;
593                 }
594                 if (flags&GRE_SEQ) {
595                         seqno = ntohl(*(__be32*)(h + offset));
596                         offset += 4;
597                 }
598         }
599
600         read_lock(&ipgre_lock);
601         if ((tunnel = ipgre_tunnel_lookup(iph->saddr, iph->daddr, key)) != NULL) {
602                 secpath_reset(skb);
603
604                 skb->protocol = *(__be16*)(h + 2);
605                 /* WCCP version 1 and 2 protocol decoding.
606                  * - Change protocol to IP
607                  * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
608                  */
609                 if (flags == 0 &&
610                     skb->protocol == htons(ETH_P_WCCP)) {
611                         skb->protocol = htons(ETH_P_IP);
612                         if ((*(h + offset) & 0xF0) != 0x40)
613                                 offset += 4;
614                 }
615
616                 skb_reset_mac_header(skb);
617                 __pskb_pull(skb, offset);
618                 skb_reset_network_header(skb);
619                 skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
620                 skb->pkt_type = PACKET_HOST;
621 #ifdef CONFIG_NET_IPGRE_BROADCAST
622                 if (MULTICAST(iph->daddr)) {
623                         /* Looped back packet, drop it! */
624                         if (((struct rtable*)skb->dst)->fl.iif == 0)
625                                 goto drop;
626                         tunnel->stat.multicast++;
627                         skb->pkt_type = PACKET_BROADCAST;
628                 }
629 #endif
630
631                 if (((flags&GRE_CSUM) && csum) ||
632                     (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
633                         tunnel->stat.rx_crc_errors++;
634                         tunnel->stat.rx_errors++;
635                         goto drop;
636                 }
637                 if (tunnel->parms.i_flags&GRE_SEQ) {
638                         if (!(flags&GRE_SEQ) ||
639                             (tunnel->i_seqno && (s32)(seqno - tunnel->i_seqno) < 0)) {
640                                 tunnel->stat.rx_fifo_errors++;
641                                 tunnel->stat.rx_errors++;
642                                 goto drop;
643                         }
644                         tunnel->i_seqno = seqno + 1;
645                 }
646                 tunnel->stat.rx_packets++;
647                 tunnel->stat.rx_bytes += skb->len;
648                 skb->dev = tunnel->dev;
649                 dst_release(skb->dst);
650                 skb->dst = NULL;
651                 nf_reset(skb);
652                 ipgre_ecn_decapsulate(iph, skb);
653                 netif_rx(skb);
654                 read_unlock(&ipgre_lock);
655                 return(0);
656         }
657         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
658
659 drop:
660         read_unlock(&ipgre_lock);
661 drop_nolock:
662         kfree_skb(skb);
663         return(0);
664 }
665
666 static int ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
667 {
668         struct ip_tunnel *tunnel = netdev_priv(dev);
669         struct net_device_stats *stats = &tunnel->stat;
670         struct iphdr  *old_iph = ip_hdr(skb);
671         struct iphdr  *tiph;
672         u8     tos;
673         __be16 df;
674         struct rtable *rt;                      /* Route to the other host */
675         struct net_device *tdev;                        /* Device to other host */
676         struct iphdr  *iph;                     /* Our new IP header */
677         int    max_headroom;                    /* The extra header space needed */
678         int    gre_hlen;
679         __be32 dst;
680         int    mtu;
681
682         if (tunnel->recursion++) {
683                 tunnel->stat.collisions++;
684                 goto tx_error;
685         }
686
687         if (dev->hard_header) {
688                 gre_hlen = 0;
689                 tiph = (struct iphdr*)skb->data;
690         } else {
691                 gre_hlen = tunnel->hlen;
692                 tiph = &tunnel->parms.iph;
693         }
694
695         if ((dst = tiph->daddr) == 0) {
696                 /* NBMA tunnel */
697
698                 if (skb->dst == NULL) {
699                         tunnel->stat.tx_fifo_errors++;
700                         goto tx_error;
701                 }
702
703                 if (skb->protocol == htons(ETH_P_IP)) {
704                         rt = (struct rtable*)skb->dst;
705                         if ((dst = rt->rt_gateway) == 0)
706                                 goto tx_error_icmp;
707                 }
708 #ifdef CONFIG_IPV6
709                 else if (skb->protocol == htons(ETH_P_IPV6)) {
710                         struct in6_addr *addr6;
711                         int addr_type;
712                         struct neighbour *neigh = skb->dst->neighbour;
713
714                         if (neigh == NULL)
715                                 goto tx_error;
716
717                         addr6 = (struct in6_addr*)&neigh->primary_key;
718                         addr_type = ipv6_addr_type(addr6);
719
720                         if (addr_type == IPV6_ADDR_ANY) {
721                                 addr6 = &ipv6_hdr(skb)->daddr;
722                                 addr_type = ipv6_addr_type(addr6);
723                         }
724
725                         if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
726                                 goto tx_error_icmp;
727
728                         dst = addr6->s6_addr32[3];
729                 }
730 #endif
731                 else
732                         goto tx_error;
733         }
734
735         tos = tiph->tos;
736         if (tos&1) {
737                 if (skb->protocol == htons(ETH_P_IP))
738                         tos = old_iph->tos;
739                 tos &= ~1;
740         }
741
742         {
743                 struct flowi fl = { .oif = tunnel->parms.link,
744                                     .nl_u = { .ip4_u =
745                                               { .daddr = dst,
746                                                 .saddr = tiph->saddr,
747                                                 .tos = RT_TOS(tos) } },
748                                     .proto = IPPROTO_GRE };
749                 if (ip_route_output_key(&rt, &fl)) {
750                         tunnel->stat.tx_carrier_errors++;
751                         goto tx_error;
752                 }
753         }
754         tdev = rt->u.dst.dev;
755
756         if (tdev == dev) {
757                 ip_rt_put(rt);
758                 tunnel->stat.collisions++;
759                 goto tx_error;
760         }
761
762         df = tiph->frag_off;
763         if (df)
764                 mtu = dst_mtu(&rt->u.dst) - tunnel->hlen;
765         else
766                 mtu = skb->dst ? dst_mtu(skb->dst) : dev->mtu;
767
768         if (skb->dst)
769                 skb->dst->ops->update_pmtu(skb->dst, mtu);
770
771         if (skb->protocol == htons(ETH_P_IP)) {
772                 df |= (old_iph->frag_off&htons(IP_DF));
773
774                 if ((old_iph->frag_off&htons(IP_DF)) &&
775                     mtu < ntohs(old_iph->tot_len)) {
776                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
777                         ip_rt_put(rt);
778                         goto tx_error;
779                 }
780         }
781 #ifdef CONFIG_IPV6
782         else if (skb->protocol == htons(ETH_P_IPV6)) {
783                 struct rt6_info *rt6 = (struct rt6_info*)skb->dst;
784
785                 if (rt6 && mtu < dst_mtu(skb->dst) && mtu >= IPV6_MIN_MTU) {
786                         if ((tunnel->parms.iph.daddr && !MULTICAST(tunnel->parms.iph.daddr)) ||
787                             rt6->rt6i_dst.plen == 128) {
788                                 rt6->rt6i_flags |= RTF_MODIFIED;
789                                 skb->dst->metrics[RTAX_MTU-1] = mtu;
790                         }
791                 }
792
793                 if (mtu >= IPV6_MIN_MTU && mtu < skb->len - tunnel->hlen + gre_hlen) {
794                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, dev);
795                         ip_rt_put(rt);
796                         goto tx_error;
797                 }
798         }
799 #endif
800
801         if (tunnel->err_count > 0) {
802                 if (jiffies - tunnel->err_time < IPTUNNEL_ERR_TIMEO) {
803                         tunnel->err_count--;
804
805                         dst_link_failure(skb);
806                 } else
807                         tunnel->err_count = 0;
808         }
809
810         max_headroom = LL_RESERVED_SPACE(tdev) + gre_hlen;
811
812         if (skb_headroom(skb) < max_headroom || skb_shared(skb)||
813             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
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 = ip_hdr(skb);
827         }
828
829         skb->transport_header = skb->network_header;
830         skb_push(skb, gre_hlen);
831         skb_reset_network_header(skb);
832         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
833         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
834                               IPSKB_REROUTED);
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                     =       ip_hdr(skb);
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         ((__be16*)(iph+1))[0] = tunnel->parms.o_flags;
863         ((__be16*)(iph+1))[1] = skb->protocol;
864
865         if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
866                 __be32 *ptr = (__be32*)(((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                         *(__sum16*)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 = netdev_priv(dev);
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 = netdev_priv(dev);
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 == netdev_priv(ipgre_fb_tunnel_dev))
1007                                 goto done;
1008                         dev = t->dev;
1009                 }
1010                 unregister_netdevice(dev);
1011                 err = 0;
1012                 break;
1013
1014         default:
1015                 err = -EINVAL;
1016         }
1017
1018 done:
1019         return err;
1020 }
1021
1022 static struct net_device_stats *ipgre_tunnel_get_stats(struct net_device *dev)
1023 {
1024         return &(((struct ip_tunnel*)netdev_priv(dev))->stat);
1025 }
1026
1027 static int ipgre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1028 {
1029         struct ip_tunnel *tunnel = netdev_priv(dev);
1030         if (new_mtu < 68 || new_mtu > 0xFFF8 - tunnel->hlen)
1031                 return -EINVAL;
1032         dev->mtu = new_mtu;
1033         return 0;
1034 }
1035
1036 #ifdef CONFIG_NET_IPGRE_BROADCAST
1037 /* Nice toy. Unfortunately, useless in real life :-)
1038    It allows to construct virtual multiprotocol broadcast "LAN"
1039    over the Internet, provided multicast routing is tuned.
1040
1041
1042    I have no idea was this bicycle invented before me,
1043    so that I had to set ARPHRD_IPGRE to a random value.
1044    I have an impression, that Cisco could make something similar,
1045    but this feature is apparently missing in IOS<=11.2(8).
1046
1047    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
1048    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
1049
1050    ping -t 255 224.66.66.66
1051
1052    If nobody answers, mbone does not work.
1053
1054    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
1055    ip addr add 10.66.66.<somewhat>/24 dev Universe
1056    ifconfig Universe up
1057    ifconfig Universe add fe80::<Your_real_addr>/10
1058    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
1059    ftp 10.66.66.66
1060    ...
1061    ftp fec0:6666:6666::193.233.7.65
1062    ...
1063
1064  */
1065
1066 static int ipgre_header(struct sk_buff *skb, struct net_device *dev, unsigned short type,
1067                         void *daddr, void *saddr, unsigned len)
1068 {
1069         struct ip_tunnel *t = netdev_priv(dev);
1070         struct iphdr *iph = (struct iphdr *)skb_push(skb, t->hlen);
1071         __be16 *p = (__be16*)(iph+1);
1072
1073         memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
1074         p[0]            = t->parms.o_flags;
1075         p[1]            = htons(type);
1076
1077         /*
1078          *      Set the source hardware address.
1079          */
1080
1081         if (saddr)
1082                 memcpy(&iph->saddr, saddr, 4);
1083
1084         if (daddr) {
1085                 memcpy(&iph->daddr, daddr, 4);
1086                 return t->hlen;
1087         }
1088         if (iph->daddr && !MULTICAST(iph->daddr))
1089                 return t->hlen;
1090
1091         return -t->hlen;
1092 }
1093
1094 static int ipgre_open(struct net_device *dev)
1095 {
1096         struct ip_tunnel *t = netdev_priv(dev);
1097
1098         if (MULTICAST(t->parms.iph.daddr)) {
1099                 struct flowi fl = { .oif = t->parms.link,
1100                                     .nl_u = { .ip4_u =
1101                                               { .daddr = t->parms.iph.daddr,
1102                                                 .saddr = t->parms.iph.saddr,
1103                                                 .tos = RT_TOS(t->parms.iph.tos) } },
1104                                     .proto = IPPROTO_GRE };
1105                 struct rtable *rt;
1106                 if (ip_route_output_key(&rt, &fl))
1107                         return -EADDRNOTAVAIL;
1108                 dev = rt->u.dst.dev;
1109                 ip_rt_put(rt);
1110                 if (__in_dev_get_rtnl(dev) == NULL)
1111                         return -EADDRNOTAVAIL;
1112                 t->mlink = dev->ifindex;
1113                 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
1114         }
1115         return 0;
1116 }
1117
1118 static int ipgre_close(struct net_device *dev)
1119 {
1120         struct ip_tunnel *t = netdev_priv(dev);
1121         if (MULTICAST(t->parms.iph.daddr) && t->mlink) {
1122                 struct in_device *in_dev = inetdev_by_index(t->mlink);
1123                 if (in_dev) {
1124                         ip_mc_dec_group(in_dev, t->parms.iph.daddr);
1125                         in_dev_put(in_dev);
1126                 }
1127         }
1128         return 0;
1129 }
1130
1131 #endif
1132
1133 static void ipgre_tunnel_setup(struct net_device *dev)
1134 {
1135         SET_MODULE_OWNER(dev);
1136         dev->uninit             = ipgre_tunnel_uninit;
1137         dev->destructor         = free_netdev;
1138         dev->hard_start_xmit    = ipgre_tunnel_xmit;
1139         dev->get_stats          = ipgre_tunnel_get_stats;
1140         dev->do_ioctl           = ipgre_tunnel_ioctl;
1141         dev->change_mtu         = ipgre_tunnel_change_mtu;
1142
1143         dev->type               = ARPHRD_IPGRE;
1144         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr) + 4;
1145         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 4;
1146         dev->flags              = IFF_NOARP;
1147         dev->iflink             = 0;
1148         dev->addr_len           = 4;
1149 }
1150
1151 static int ipgre_tunnel_init(struct net_device *dev)
1152 {
1153         struct net_device *tdev = NULL;
1154         struct ip_tunnel *tunnel;
1155         struct iphdr *iph;
1156         int hlen = LL_MAX_HEADER;
1157         int mtu = ETH_DATA_LEN;
1158         int addend = sizeof(struct iphdr) + 4;
1159
1160         tunnel = netdev_priv(dev);
1161         iph = &tunnel->parms.iph;
1162
1163         tunnel->dev = dev;
1164         strcpy(tunnel->parms.name, dev->name);
1165
1166         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1167         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1168
1169         /* Guess output device to choose reasonable mtu and hard_header_len */
1170
1171         if (iph->daddr) {
1172                 struct flowi fl = { .oif = tunnel->parms.link,
1173                                     .nl_u = { .ip4_u =
1174                                               { .daddr = iph->daddr,
1175                                                 .saddr = iph->saddr,
1176                                                 .tos = RT_TOS(iph->tos) } },
1177                                     .proto = IPPROTO_GRE };
1178                 struct rtable *rt;
1179                 if (!ip_route_output_key(&rt, &fl)) {
1180                         tdev = rt->u.dst.dev;
1181                         ip_rt_put(rt);
1182                 }
1183
1184                 dev->flags |= IFF_POINTOPOINT;
1185
1186 #ifdef CONFIG_NET_IPGRE_BROADCAST
1187                 if (MULTICAST(iph->daddr)) {
1188                         if (!iph->saddr)
1189                                 return -EINVAL;
1190                         dev->flags = IFF_BROADCAST;
1191                         dev->hard_header = ipgre_header;
1192                         dev->open = ipgre_open;
1193                         dev->stop = ipgre_close;
1194                 }
1195 #endif
1196         }
1197
1198         if (!tdev && tunnel->parms.link)
1199                 tdev = __dev_get_by_index(tunnel->parms.link);
1200
1201         if (tdev) {
1202                 hlen = tdev->hard_header_len;
1203                 mtu = tdev->mtu;
1204         }
1205         dev->iflink = tunnel->parms.link;
1206
1207         /* Precalculate GRE options length */
1208         if (tunnel->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
1209                 if (tunnel->parms.o_flags&GRE_CSUM)
1210                         addend += 4;
1211                 if (tunnel->parms.o_flags&GRE_KEY)
1212                         addend += 4;
1213                 if (tunnel->parms.o_flags&GRE_SEQ)
1214                         addend += 4;
1215         }
1216         dev->hard_header_len = hlen + addend;
1217         dev->mtu = mtu - addend;
1218         tunnel->hlen = addend;
1219         return 0;
1220 }
1221
1222 static int __init ipgre_fb_tunnel_init(struct net_device *dev)
1223 {
1224         struct ip_tunnel *tunnel = netdev_priv(dev);
1225         struct iphdr *iph = &tunnel->parms.iph;
1226
1227         tunnel->dev = dev;
1228         strcpy(tunnel->parms.name, dev->name);
1229
1230         iph->version            = 4;
1231         iph->protocol           = IPPROTO_GRE;
1232         iph->ihl                = 5;
1233         tunnel->hlen            = sizeof(struct iphdr) + 4;
1234
1235         dev_hold(dev);
1236         tunnels_wc[0]           = tunnel;
1237         return 0;
1238 }
1239
1240
1241 static struct net_protocol ipgre_protocol = {
1242         .handler        =       ipgre_rcv,
1243         .err_handler    =       ipgre_err,
1244 };
1245
1246
1247 /*
1248  *      And now the modules code and kernel interface.
1249  */
1250
1251 static int __init ipgre_init(void)
1252 {
1253         int err;
1254
1255         printk(KERN_INFO "GRE over IPv4 tunneling driver\n");
1256
1257         if (inet_add_protocol(&ipgre_protocol, IPPROTO_GRE) < 0) {
1258                 printk(KERN_INFO "ipgre init: can't add protocol\n");
1259                 return -EAGAIN;
1260         }
1261
1262         ipgre_fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "gre0",
1263                                            ipgre_tunnel_setup);
1264         if (!ipgre_fb_tunnel_dev) {
1265                 err = -ENOMEM;
1266                 goto err1;
1267         }
1268
1269         ipgre_fb_tunnel_dev->init = ipgre_fb_tunnel_init;
1270
1271         if ((err = register_netdev(ipgre_fb_tunnel_dev)))
1272                 goto err2;
1273 out:
1274         return err;
1275 err2:
1276         free_netdev(ipgre_fb_tunnel_dev);
1277 err1:
1278         inet_del_protocol(&ipgre_protocol, IPPROTO_GRE);
1279         goto out;
1280 }
1281
1282 static void __exit ipgre_destroy_tunnels(void)
1283 {
1284         int prio;
1285
1286         for (prio = 0; prio < 4; prio++) {
1287                 int h;
1288                 for (h = 0; h < HASH_SIZE; h++) {
1289                         struct ip_tunnel *t;
1290                         while ((t = tunnels[prio][h]) != NULL)
1291                                 unregister_netdevice(t->dev);
1292                 }
1293         }
1294 }
1295
1296 static void __exit ipgre_fini(void)
1297 {
1298         if (inet_del_protocol(&ipgre_protocol, IPPROTO_GRE) < 0)
1299                 printk(KERN_INFO "ipgre close: can't remove protocol\n");
1300
1301         rtnl_lock();
1302         ipgre_destroy_tunnels();
1303         rtnl_unlock();
1304 }
1305
1306 module_init(ipgre_init);
1307 module_exit(ipgre_fini);
1308 MODULE_LICENSE("GPL");