[SK_BUFF]: Introduce ip_hdr(), remove skb->nh.iph
[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 *t)
195 {
196         __be32 remote = t->parms.iph.daddr;
197         __be32 local = t->parms.iph.saddr;
198         __be32 key = t->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 void ipgre_tunnel_link(struct ip_tunnel *t)
213 {
214         struct ip_tunnel **tp = ipgre_bucket(t);
215
216         t->next = *tp;
217         write_lock_bh(&ipgre_lock);
218         *tp = t;
219         write_unlock_bh(&ipgre_lock);
220 }
221
222 static void ipgre_tunnel_unlink(struct ip_tunnel *t)
223 {
224         struct ip_tunnel **tp;
225
226         for (tp = ipgre_bucket(t); *tp; tp = &(*tp)->next) {
227                 if (t == *tp) {
228                         write_lock_bh(&ipgre_lock);
229                         *tp = t->next;
230                         write_unlock_bh(&ipgre_lock);
231                         break;
232                 }
233         }
234 }
235
236 static struct ip_tunnel * ipgre_tunnel_locate(struct ip_tunnel_parm *parms, int create)
237 {
238         __be32 remote = parms->iph.daddr;
239         __be32 local = parms->iph.saddr;
240         __be32 key = parms->i_key;
241         struct ip_tunnel *t, **tp, *nt;
242         struct net_device *dev;
243         unsigned h = HASH(key);
244         int prio = 0;
245         char name[IFNAMSIZ];
246
247         if (local)
248                 prio |= 1;
249         if (remote && !MULTICAST(remote)) {
250                 prio |= 2;
251                 h ^= HASH(remote);
252         }
253         for (tp = &tunnels[prio][h]; (t = *tp) != NULL; tp = &t->next) {
254                 if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr) {
255                         if (key == t->parms.i_key)
256                                 return t;
257                 }
258         }
259         if (!create)
260                 return NULL;
261
262         if (parms->name[0])
263                 strlcpy(name, parms->name, IFNAMSIZ);
264         else {
265                 int i;
266                 for (i=1; i<100; i++) {
267                         sprintf(name, "gre%d", i);
268                         if (__dev_get_by_name(name) == NULL)
269                                 break;
270                 }
271                 if (i==100)
272                         goto failed;
273         }
274
275         dev = alloc_netdev(sizeof(*t), name, ipgre_tunnel_setup);
276         if (!dev)
277           return NULL;
278
279         dev->init = ipgre_tunnel_init;
280         nt = netdev_priv(dev);
281         nt->parms = *parms;
282
283         if (register_netdevice(dev) < 0) {
284                 free_netdev(dev);
285                 goto failed;
286         }
287
288         dev_hold(dev);
289         ipgre_tunnel_link(nt);
290         return nt;
291
292 failed:
293         return NULL;
294 }
295
296 static void ipgre_tunnel_uninit(struct net_device *dev)
297 {
298         ipgre_tunnel_unlink(netdev_priv(dev));
299         dev_put(dev);
300 }
301
302
303 static void ipgre_err(struct sk_buff *skb, u32 info)
304 {
305 #ifndef I_WISH_WORLD_WERE_PERFECT
306
307 /* It is not :-( All the routers (except for Linux) return only
308    8 bytes of packet payload. It means, that precise relaying of
309    ICMP in the real Internet is absolutely infeasible.
310
311    Moreover, Cisco "wise men" put GRE key to the third word
312    in GRE header. It makes impossible maintaining even soft state for keyed
313    GRE tunnels with enabled checksum. Tell them "thank you".
314
315    Well, I wonder, rfc1812 was written by Cisco employee,
316    what the hell these idiots break standrads established
317    by themself???
318  */
319
320         struct iphdr *iph = (struct iphdr*)skb->data;
321         __be16       *p = (__be16*)(skb->data+(iph->ihl<<2));
322         int grehlen = (iph->ihl<<2) + 4;
323         int type = skb->h.icmph->type;
324         int code = skb->h.icmph->code;
325         struct ip_tunnel *t;
326         __be16 flags;
327
328         flags = p[0];
329         if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
330                 if (flags&(GRE_VERSION|GRE_ROUTING))
331                         return;
332                 if (flags&GRE_KEY) {
333                         grehlen += 4;
334                         if (flags&GRE_CSUM)
335                                 grehlen += 4;
336                 }
337         }
338
339         /* If only 8 bytes returned, keyed message will be dropped here */
340         if (skb_headlen(skb) < grehlen)
341                 return;
342
343         switch (type) {
344         default:
345         case ICMP_PARAMETERPROB:
346                 return;
347
348         case ICMP_DEST_UNREACH:
349                 switch (code) {
350                 case ICMP_SR_FAILED:
351                 case ICMP_PORT_UNREACH:
352                         /* Impossible event. */
353                         return;
354                 case ICMP_FRAG_NEEDED:
355                         /* Soft state for pmtu is maintained by IP core. */
356                         return;
357                 default:
358                         /* All others are translated to HOST_UNREACH.
359                            rfc2003 contains "deep thoughts" about NET_UNREACH,
360                            I believe they are just ether pollution. --ANK
361                          */
362                         break;
363                 }
364                 break;
365         case ICMP_TIME_EXCEEDED:
366                 if (code != ICMP_EXC_TTL)
367                         return;
368                 break;
369         }
370
371         read_lock(&ipgre_lock);
372         t = ipgre_tunnel_lookup(iph->daddr, iph->saddr, (flags&GRE_KEY) ? *(((__be32*)p) + (grehlen>>2) - 1) : 0);
373         if (t == NULL || t->parms.iph.daddr == 0 || MULTICAST(t->parms.iph.daddr))
374                 goto out;
375
376         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
377                 goto out;
378
379         if (jiffies - t->err_time < IPTUNNEL_ERR_TIMEO)
380                 t->err_count++;
381         else
382                 t->err_count = 1;
383         t->err_time = jiffies;
384 out:
385         read_unlock(&ipgre_lock);
386         return;
387 #else
388         struct iphdr *iph = (struct iphdr*)dp;
389         struct iphdr *eiph;
390         __be16       *p = (__be16*)(dp+(iph->ihl<<2));
391         int type = skb->h.icmph->type;
392         int code = skb->h.icmph->code;
393         int rel_type = 0;
394         int rel_code = 0;
395         __be32 rel_info = 0;
396         __u32 n = 0;
397         __be16 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                 n = ntohl(skb->h.icmph->un.gateway) >> 24;
426                 if (n < (iph->ihl<<2))
427                         return;
428
429                 /* So... This guy found something strange INSIDE encapsulated
430                    packet. Well, he is fool, but what can we do ?
431                  */
432                 rel_type = ICMP_PARAMETERPROB;
433                 n -= grehlen;
434                 rel_info = htonl(n << 24);
435                 break;
436
437         case ICMP_DEST_UNREACH:
438                 switch (code) {
439                 case ICMP_SR_FAILED:
440                 case ICMP_PORT_UNREACH:
441                         /* Impossible event. */
442                         return;
443                 case ICMP_FRAG_NEEDED:
444                         /* And it is the only really necessary thing :-) */
445                         n = ntohs(skb->h.icmph->un.frag.mtu);
446                         if (n < grehlen+68)
447                                 return;
448                         n -= grehlen;
449                         /* BSD 4.2 MORE DOES NOT EXIST IN NATURE. */
450                         if (n > ntohs(eiph->tot_len))
451                                 return;
452                         rel_info = htonl(n);
453                         break;
454                 default:
455                         /* All others are translated to HOST_UNREACH.
456                            rfc2003 contains "deep thoughts" about NET_UNREACH,
457                            I believe, it is just ether pollution. --ANK
458                          */
459                         rel_type = ICMP_DEST_UNREACH;
460                         rel_code = ICMP_HOST_UNREACH;
461                         break;
462                 }
463                 break;
464         case ICMP_TIME_EXCEEDED:
465                 if (code != ICMP_EXC_TTL)
466                         return;
467                 break;
468         }
469
470         /* Prepare fake skb to feed it to icmp_send */
471         skb2 = skb_clone(skb, GFP_ATOMIC);
472         if (skb2 == NULL)
473                 return;
474         dst_release(skb2->dst);
475         skb2->dst = NULL;
476         skb_pull(skb2, skb->data - (u8*)eiph);
477         skb_reset_network_header(skb2);
478
479         /* Try to guess incoming interface */
480         memset(&fl, 0, sizeof(fl));
481         fl.fl4_dst = eiph->saddr;
482         fl.fl4_tos = RT_TOS(eiph->tos);
483         fl.proto = IPPROTO_GRE;
484         if (ip_route_output_key(&rt, &fl)) {
485                 kfree_skb(skb2);
486                 return;
487         }
488         skb2->dev = rt->u.dst.dev;
489
490         /* route "incoming" packet */
491         if (rt->rt_flags&RTCF_LOCAL) {
492                 ip_rt_put(rt);
493                 rt = NULL;
494                 fl.fl4_dst = eiph->daddr;
495                 fl.fl4_src = eiph->saddr;
496                 fl.fl4_tos = eiph->tos;
497                 if (ip_route_output_key(&rt, &fl) ||
498                     rt->u.dst.dev->type != ARPHRD_IPGRE) {
499                         ip_rt_put(rt);
500                         kfree_skb(skb2);
501                         return;
502                 }
503         } else {
504                 ip_rt_put(rt);
505                 if (ip_route_input(skb2, eiph->daddr, eiph->saddr, eiph->tos, skb2->dev) ||
506                     skb2->dst->dev->type != ARPHRD_IPGRE) {
507                         kfree_skb(skb2);
508                         return;
509                 }
510         }
511
512         /* change mtu on this route */
513         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
514                 if (n > dst_mtu(skb2->dst)) {
515                         kfree_skb(skb2);
516                         return;
517                 }
518                 skb2->dst->ops->update_pmtu(skb2->dst, n);
519         } else if (type == ICMP_TIME_EXCEEDED) {
520                 struct ip_tunnel *t = netdev_priv(skb2->dev);
521                 if (t->parms.iph.ttl) {
522                         rel_type = ICMP_DEST_UNREACH;
523                         rel_code = ICMP_HOST_UNREACH;
524                 }
525         }
526
527         icmp_send(skb2, rel_type, rel_code, rel_info);
528         kfree_skb(skb2);
529 #endif
530 }
531
532 static inline void ipgre_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb)
533 {
534         if (INET_ECN_is_ce(iph->tos)) {
535                 if (skb->protocol == htons(ETH_P_IP)) {
536                         IP_ECN_set_ce(ip_hdr(skb));
537                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
538                         IP6_ECN_set_ce(skb->nh.ipv6h);
539                 }
540         }
541 }
542
543 static inline u8
544 ipgre_ecn_encapsulate(u8 tos, struct iphdr *old_iph, struct sk_buff *skb)
545 {
546         u8 inner = 0;
547         if (skb->protocol == htons(ETH_P_IP))
548                 inner = old_iph->tos;
549         else if (skb->protocol == htons(ETH_P_IPV6))
550                 inner = ipv6_get_dsfield((struct ipv6hdr *)old_iph);
551         return INET_ECN_encapsulate(tos, inner);
552 }
553
554 static int ipgre_rcv(struct sk_buff *skb)
555 {
556         struct iphdr *iph;
557         u8     *h;
558         __be16    flags;
559         __sum16   csum = 0;
560         __be32 key = 0;
561         u32    seqno = 0;
562         struct ip_tunnel *tunnel;
563         int    offset = 4;
564
565         if (!pskb_may_pull(skb, 16))
566                 goto drop_nolock;
567
568         iph = ip_hdr(skb);
569         h = skb->data;
570         flags = *(__be16*)h;
571
572         if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
573                 /* - Version must be 0.
574                    - We do not support routing headers.
575                  */
576                 if (flags&(GRE_VERSION|GRE_ROUTING))
577                         goto drop_nolock;
578
579                 if (flags&GRE_CSUM) {
580                         switch (skb->ip_summed) {
581                         case CHECKSUM_COMPLETE:
582                                 csum = csum_fold(skb->csum);
583                                 if (!csum)
584                                         break;
585                                 /* fall through */
586                         case CHECKSUM_NONE:
587                                 skb->csum = 0;
588                                 csum = __skb_checksum_complete(skb);
589                                 skb->ip_summed = CHECKSUM_COMPLETE;
590                         }
591                         offset += 4;
592                 }
593                 if (flags&GRE_KEY) {
594                         key = *(__be32*)(h + offset);
595                         offset += 4;
596                 }
597                 if (flags&GRE_SEQ) {
598                         seqno = ntohl(*(__be32*)(h + offset));
599                         offset += 4;
600                 }
601         }
602
603         read_lock(&ipgre_lock);
604         if ((tunnel = ipgre_tunnel_lookup(iph->saddr, iph->daddr, key)) != NULL) {
605                 secpath_reset(skb);
606
607                 skb->protocol = *(__be16*)(h + 2);
608                 /* WCCP version 1 and 2 protocol decoding.
609                  * - Change protocol to IP
610                  * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
611                  */
612                 if (flags == 0 &&
613                     skb->protocol == htons(ETH_P_WCCP)) {
614                         skb->protocol = htons(ETH_P_IP);
615                         if ((*(h + offset) & 0xF0) != 0x40)
616                                 offset += 4;
617                 }
618
619                 skb_reset_mac_header(skb);
620                 __pskb_pull(skb, offset);
621                 skb_reset_network_header(skb);
622                 skb_postpull_rcsum(skb, skb->h.raw, offset);
623                 skb->pkt_type = PACKET_HOST;
624 #ifdef CONFIG_NET_IPGRE_BROADCAST
625                 if (MULTICAST(iph->daddr)) {
626                         /* Looped back packet, drop it! */
627                         if (((struct rtable*)skb->dst)->fl.iif == 0)
628                                 goto drop;
629                         tunnel->stat.multicast++;
630                         skb->pkt_type = PACKET_BROADCAST;
631                 }
632 #endif
633
634                 if (((flags&GRE_CSUM) && csum) ||
635                     (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
636                         tunnel->stat.rx_crc_errors++;
637                         tunnel->stat.rx_errors++;
638                         goto drop;
639                 }
640                 if (tunnel->parms.i_flags&GRE_SEQ) {
641                         if (!(flags&GRE_SEQ) ||
642                             (tunnel->i_seqno && (s32)(seqno - tunnel->i_seqno) < 0)) {
643                                 tunnel->stat.rx_fifo_errors++;
644                                 tunnel->stat.rx_errors++;
645                                 goto drop;
646                         }
647                         tunnel->i_seqno = seqno + 1;
648                 }
649                 tunnel->stat.rx_packets++;
650                 tunnel->stat.rx_bytes += skb->len;
651                 skb->dev = tunnel->dev;
652                 dst_release(skb->dst);
653                 skb->dst = NULL;
654                 nf_reset(skb);
655                 ipgre_ecn_decapsulate(iph, skb);
656                 netif_rx(skb);
657                 read_unlock(&ipgre_lock);
658                 return(0);
659         }
660         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
661
662 drop:
663         read_unlock(&ipgre_lock);
664 drop_nolock:
665         kfree_skb(skb);
666         return(0);
667 }
668
669 static int ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
670 {
671         struct ip_tunnel *tunnel = netdev_priv(dev);
672         struct net_device_stats *stats = &tunnel->stat;
673         struct iphdr  *old_iph = ip_hdr(skb);
674         struct iphdr  *tiph;
675         u8     tos;
676         __be16 df;
677         struct rtable *rt;                      /* Route to the other host */
678         struct net_device *tdev;                        /* Device to other host */
679         struct iphdr  *iph;                     /* Our new IP header */
680         int    max_headroom;                    /* The extra header space needed */
681         int    gre_hlen;
682         __be32 dst;
683         int    mtu;
684
685         if (tunnel->recursion++) {
686                 tunnel->stat.collisions++;
687                 goto tx_error;
688         }
689
690         if (dev->hard_header) {
691                 gre_hlen = 0;
692                 tiph = (struct iphdr*)skb->data;
693         } else {
694                 gre_hlen = tunnel->hlen;
695                 tiph = &tunnel->parms.iph;
696         }
697
698         if ((dst = tiph->daddr) == 0) {
699                 /* NBMA tunnel */
700
701                 if (skb->dst == NULL) {
702                         tunnel->stat.tx_fifo_errors++;
703                         goto tx_error;
704                 }
705
706                 if (skb->protocol == htons(ETH_P_IP)) {
707                         rt = (struct rtable*)skb->dst;
708                         if ((dst = rt->rt_gateway) == 0)
709                                 goto tx_error_icmp;
710                 }
711 #ifdef CONFIG_IPV6
712                 else if (skb->protocol == htons(ETH_P_IPV6)) {
713                         struct in6_addr *addr6;
714                         int addr_type;
715                         struct neighbour *neigh = skb->dst->neighbour;
716
717                         if (neigh == NULL)
718                                 goto tx_error;
719
720                         addr6 = (struct in6_addr*)&neigh->primary_key;
721                         addr_type = ipv6_addr_type(addr6);
722
723                         if (addr_type == IPV6_ADDR_ANY) {
724                                 addr6 = &skb->nh.ipv6h->daddr;
725                                 addr_type = ipv6_addr_type(addr6);
726                         }
727
728                         if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
729                                 goto tx_error_icmp;
730
731                         dst = addr6->s6_addr32[3];
732                 }
733 #endif
734                 else
735                         goto tx_error;
736         }
737
738         tos = tiph->tos;
739         if (tos&1) {
740                 if (skb->protocol == htons(ETH_P_IP))
741                         tos = old_iph->tos;
742                 tos &= ~1;
743         }
744
745         {
746                 struct flowi fl = { .oif = tunnel->parms.link,
747                                     .nl_u = { .ip4_u =
748                                               { .daddr = dst,
749                                                 .saddr = tiph->saddr,
750                                                 .tos = RT_TOS(tos) } },
751                                     .proto = IPPROTO_GRE };
752                 if (ip_route_output_key(&rt, &fl)) {
753                         tunnel->stat.tx_carrier_errors++;
754                         goto tx_error;
755                 }
756         }
757         tdev = rt->u.dst.dev;
758
759         if (tdev == dev) {
760                 ip_rt_put(rt);
761                 tunnel->stat.collisions++;
762                 goto tx_error;
763         }
764
765         df = tiph->frag_off;
766         if (df)
767                 mtu = dst_mtu(&rt->u.dst) - tunnel->hlen;
768         else
769                 mtu = skb->dst ? dst_mtu(skb->dst) : dev->mtu;
770
771         if (skb->dst)
772                 skb->dst->ops->update_pmtu(skb->dst, mtu);
773
774         if (skb->protocol == htons(ETH_P_IP)) {
775                 df |= (old_iph->frag_off&htons(IP_DF));
776
777                 if ((old_iph->frag_off&htons(IP_DF)) &&
778                     mtu < ntohs(old_iph->tot_len)) {
779                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
780                         ip_rt_put(rt);
781                         goto tx_error;
782                 }
783         }
784 #ifdef CONFIG_IPV6
785         else if (skb->protocol == htons(ETH_P_IPV6)) {
786                 struct rt6_info *rt6 = (struct rt6_info*)skb->dst;
787
788                 if (rt6 && mtu < dst_mtu(skb->dst) && mtu >= IPV6_MIN_MTU) {
789                         if ((tunnel->parms.iph.daddr && !MULTICAST(tunnel->parms.iph.daddr)) ||
790                             rt6->rt6i_dst.plen == 128) {
791                                 rt6->rt6i_flags |= RTF_MODIFIED;
792                                 skb->dst->metrics[RTAX_MTU-1] = mtu;
793                         }
794                 }
795
796                 if (mtu >= IPV6_MIN_MTU && mtu < skb->len - tunnel->hlen + gre_hlen) {
797                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, dev);
798                         ip_rt_put(rt);
799                         goto tx_error;
800                 }
801         }
802 #endif
803
804         if (tunnel->err_count > 0) {
805                 if (jiffies - tunnel->err_time < IPTUNNEL_ERR_TIMEO) {
806                         tunnel->err_count--;
807
808                         dst_link_failure(skb);
809                 } else
810                         tunnel->err_count = 0;
811         }
812
813         max_headroom = LL_RESERVED_SPACE(tdev) + gre_hlen;
814
815         if (skb_headroom(skb) < max_headroom || skb_cloned(skb) || skb_shared(skb)) {
816                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
817                 if (!new_skb) {
818                         ip_rt_put(rt);
819                         stats->tx_dropped++;
820                         dev_kfree_skb(skb);
821                         tunnel->recursion--;
822                         return 0;
823                 }
824                 if (skb->sk)
825                         skb_set_owner_w(new_skb, skb->sk);
826                 dev_kfree_skb(skb);
827                 skb = new_skb;
828                 old_iph = ip_hdr(skb);
829         }
830
831         skb->h.raw = skb->nh.raw;
832         skb_push(skb, gre_hlen);
833         skb_reset_network_header(skb);
834         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
835         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
836                               IPSKB_REROUTED);
837         dst_release(skb->dst);
838         skb->dst = &rt->u.dst;
839
840         /*
841          *      Push down and install the IPIP header.
842          */
843
844         iph                     =       ip_hdr(skb);
845         iph->version            =       4;
846         iph->ihl                =       sizeof(struct iphdr) >> 2;
847         iph->frag_off           =       df;
848         iph->protocol           =       IPPROTO_GRE;
849         iph->tos                =       ipgre_ecn_encapsulate(tos, old_iph, skb);
850         iph->daddr              =       rt->rt_dst;
851         iph->saddr              =       rt->rt_src;
852
853         if ((iph->ttl = tiph->ttl) == 0) {
854                 if (skb->protocol == htons(ETH_P_IP))
855                         iph->ttl = old_iph->ttl;
856 #ifdef CONFIG_IPV6
857                 else if (skb->protocol == htons(ETH_P_IPV6))
858                         iph->ttl = ((struct ipv6hdr*)old_iph)->hop_limit;
859 #endif
860                 else
861                         iph->ttl = dst_metric(&rt->u.dst, RTAX_HOPLIMIT);
862         }
863
864         ((__be16*)(iph+1))[0] = tunnel->parms.o_flags;
865         ((__be16*)(iph+1))[1] = skb->protocol;
866
867         if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
868                 __be32 *ptr = (__be32*)(((u8*)iph) + tunnel->hlen - 4);
869
870                 if (tunnel->parms.o_flags&GRE_SEQ) {
871                         ++tunnel->o_seqno;
872                         *ptr = htonl(tunnel->o_seqno);
873                         ptr--;
874                 }
875                 if (tunnel->parms.o_flags&GRE_KEY) {
876                         *ptr = tunnel->parms.o_key;
877                         ptr--;
878                 }
879                 if (tunnel->parms.o_flags&GRE_CSUM) {
880                         *ptr = 0;
881                         *(__sum16*)ptr = ip_compute_csum((void*)(iph+1), skb->len - sizeof(struct iphdr));
882                 }
883         }
884
885         nf_reset(skb);
886
887         IPTUNNEL_XMIT();
888         tunnel->recursion--;
889         return 0;
890
891 tx_error_icmp:
892         dst_link_failure(skb);
893
894 tx_error:
895         stats->tx_errors++;
896         dev_kfree_skb(skb);
897         tunnel->recursion--;
898         return 0;
899 }
900
901 static int
902 ipgre_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
903 {
904         int err = 0;
905         struct ip_tunnel_parm p;
906         struct ip_tunnel *t;
907
908         switch (cmd) {
909         case SIOCGETTUNNEL:
910                 t = NULL;
911                 if (dev == ipgre_fb_tunnel_dev) {
912                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
913                                 err = -EFAULT;
914                                 break;
915                         }
916                         t = ipgre_tunnel_locate(&p, 0);
917                 }
918                 if (t == NULL)
919                         t = netdev_priv(dev);
920                 memcpy(&p, &t->parms, sizeof(p));
921                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
922                         err = -EFAULT;
923                 break;
924
925         case SIOCADDTUNNEL:
926         case SIOCCHGTUNNEL:
927                 err = -EPERM;
928                 if (!capable(CAP_NET_ADMIN))
929                         goto done;
930
931                 err = -EFAULT;
932                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
933                         goto done;
934
935                 err = -EINVAL;
936                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
937                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) ||
938                     ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING)))
939                         goto done;
940                 if (p.iph.ttl)
941                         p.iph.frag_off |= htons(IP_DF);
942
943                 if (!(p.i_flags&GRE_KEY))
944                         p.i_key = 0;
945                 if (!(p.o_flags&GRE_KEY))
946                         p.o_key = 0;
947
948                 t = ipgre_tunnel_locate(&p, cmd == SIOCADDTUNNEL);
949
950                 if (dev != ipgre_fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
951                         if (t != NULL) {
952                                 if (t->dev != dev) {
953                                         err = -EEXIST;
954                                         break;
955                                 }
956                         } else {
957                                 unsigned nflags=0;
958
959                                 t = netdev_priv(dev);
960
961                                 if (MULTICAST(p.iph.daddr))
962                                         nflags = IFF_BROADCAST;
963                                 else if (p.iph.daddr)
964                                         nflags = IFF_POINTOPOINT;
965
966                                 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) {
967                                         err = -EINVAL;
968                                         break;
969                                 }
970                                 ipgre_tunnel_unlink(t);
971                                 t->parms.iph.saddr = p.iph.saddr;
972                                 t->parms.iph.daddr = p.iph.daddr;
973                                 t->parms.i_key = p.i_key;
974                                 t->parms.o_key = p.o_key;
975                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
976                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
977                                 ipgre_tunnel_link(t);
978                                 netdev_state_change(dev);
979                         }
980                 }
981
982                 if (t) {
983                         err = 0;
984                         if (cmd == SIOCCHGTUNNEL) {
985                                 t->parms.iph.ttl = p.iph.ttl;
986                                 t->parms.iph.tos = p.iph.tos;
987                                 t->parms.iph.frag_off = p.iph.frag_off;
988                         }
989                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
990                                 err = -EFAULT;
991                 } else
992                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
993                 break;
994
995         case SIOCDELTUNNEL:
996                 err = -EPERM;
997                 if (!capable(CAP_NET_ADMIN))
998                         goto done;
999
1000                 if (dev == ipgre_fb_tunnel_dev) {
1001                         err = -EFAULT;
1002                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1003                                 goto done;
1004                         err = -ENOENT;
1005                         if ((t = ipgre_tunnel_locate(&p, 0)) == NULL)
1006                                 goto done;
1007                         err = -EPERM;
1008                         if (t == netdev_priv(ipgre_fb_tunnel_dev))
1009                                 goto done;
1010                         dev = t->dev;
1011                 }
1012                 unregister_netdevice(dev);
1013                 err = 0;
1014                 break;
1015
1016         default:
1017                 err = -EINVAL;
1018         }
1019
1020 done:
1021         return err;
1022 }
1023
1024 static struct net_device_stats *ipgre_tunnel_get_stats(struct net_device *dev)
1025 {
1026         return &(((struct ip_tunnel*)netdev_priv(dev))->stat);
1027 }
1028
1029 static int ipgre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1030 {
1031         struct ip_tunnel *tunnel = netdev_priv(dev);
1032         if (new_mtu < 68 || new_mtu > 0xFFF8 - tunnel->hlen)
1033                 return -EINVAL;
1034         dev->mtu = new_mtu;
1035         return 0;
1036 }
1037
1038 #ifdef CONFIG_NET_IPGRE_BROADCAST
1039 /* Nice toy. Unfortunately, useless in real life :-)
1040    It allows to construct virtual multiprotocol broadcast "LAN"
1041    over the Internet, provided multicast routing is tuned.
1042
1043
1044    I have no idea was this bicycle invented before me,
1045    so that I had to set ARPHRD_IPGRE to a random value.
1046    I have an impression, that Cisco could make something similar,
1047    but this feature is apparently missing in IOS<=11.2(8).
1048
1049    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
1050    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
1051
1052    ping -t 255 224.66.66.66
1053
1054    If nobody answers, mbone does not work.
1055
1056    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
1057    ip addr add 10.66.66.<somewhat>/24 dev Universe
1058    ifconfig Universe up
1059    ifconfig Universe add fe80::<Your_real_addr>/10
1060    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
1061    ftp 10.66.66.66
1062    ...
1063    ftp fec0:6666:6666::193.233.7.65
1064    ...
1065
1066  */
1067
1068 static int ipgre_header(struct sk_buff *skb, struct net_device *dev, unsigned short type,
1069                         void *daddr, void *saddr, unsigned len)
1070 {
1071         struct ip_tunnel *t = netdev_priv(dev);
1072         struct iphdr *iph = (struct iphdr *)skb_push(skb, t->hlen);
1073         __be16 *p = (__be16*)(iph+1);
1074
1075         memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
1076         p[0]            = t->parms.o_flags;
1077         p[1]            = htons(type);
1078
1079         /*
1080          *      Set the source hardware address.
1081          */
1082
1083         if (saddr)
1084                 memcpy(&iph->saddr, saddr, 4);
1085
1086         if (daddr) {
1087                 memcpy(&iph->daddr, daddr, 4);
1088                 return t->hlen;
1089         }
1090         if (iph->daddr && !MULTICAST(iph->daddr))
1091                 return t->hlen;
1092
1093         return -t->hlen;
1094 }
1095
1096 static int ipgre_open(struct net_device *dev)
1097 {
1098         struct ip_tunnel *t = netdev_priv(dev);
1099
1100         if (MULTICAST(t->parms.iph.daddr)) {
1101                 struct flowi fl = { .oif = t->parms.link,
1102                                     .nl_u = { .ip4_u =
1103                                               { .daddr = t->parms.iph.daddr,
1104                                                 .saddr = t->parms.iph.saddr,
1105                                                 .tos = RT_TOS(t->parms.iph.tos) } },
1106                                     .proto = IPPROTO_GRE };
1107                 struct rtable *rt;
1108                 if (ip_route_output_key(&rt, &fl))
1109                         return -EADDRNOTAVAIL;
1110                 dev = rt->u.dst.dev;
1111                 ip_rt_put(rt);
1112                 if (__in_dev_get_rtnl(dev) == NULL)
1113                         return -EADDRNOTAVAIL;
1114                 t->mlink = dev->ifindex;
1115                 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
1116         }
1117         return 0;
1118 }
1119
1120 static int ipgre_close(struct net_device *dev)
1121 {
1122         struct ip_tunnel *t = netdev_priv(dev);
1123         if (MULTICAST(t->parms.iph.daddr) && t->mlink) {
1124                 struct in_device *in_dev = inetdev_by_index(t->mlink);
1125                 if (in_dev) {
1126                         ip_mc_dec_group(in_dev, t->parms.iph.daddr);
1127                         in_dev_put(in_dev);
1128                 }
1129         }
1130         return 0;
1131 }
1132
1133 #endif
1134
1135 static void ipgre_tunnel_setup(struct net_device *dev)
1136 {
1137         SET_MODULE_OWNER(dev);
1138         dev->uninit             = ipgre_tunnel_uninit;
1139         dev->destructor         = free_netdev;
1140         dev->hard_start_xmit    = ipgre_tunnel_xmit;
1141         dev->get_stats          = ipgre_tunnel_get_stats;
1142         dev->do_ioctl           = ipgre_tunnel_ioctl;
1143         dev->change_mtu         = ipgre_tunnel_change_mtu;
1144
1145         dev->type               = ARPHRD_IPGRE;
1146         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr) + 4;
1147         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 4;
1148         dev->flags              = IFF_NOARP;
1149         dev->iflink             = 0;
1150         dev->addr_len           = 4;
1151 }
1152
1153 static int ipgre_tunnel_init(struct net_device *dev)
1154 {
1155         struct net_device *tdev = NULL;
1156         struct ip_tunnel *tunnel;
1157         struct iphdr *iph;
1158         int hlen = LL_MAX_HEADER;
1159         int mtu = ETH_DATA_LEN;
1160         int addend = sizeof(struct iphdr) + 4;
1161
1162         tunnel = netdev_priv(dev);
1163         iph = &tunnel->parms.iph;
1164
1165         tunnel->dev = dev;
1166         strcpy(tunnel->parms.name, dev->name);
1167
1168         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1169         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1170
1171         /* Guess output device to choose reasonable mtu and hard_header_len */
1172
1173         if (iph->daddr) {
1174                 struct flowi fl = { .oif = tunnel->parms.link,
1175                                     .nl_u = { .ip4_u =
1176                                               { .daddr = iph->daddr,
1177                                                 .saddr = iph->saddr,
1178                                                 .tos = RT_TOS(iph->tos) } },
1179                                     .proto = IPPROTO_GRE };
1180                 struct rtable *rt;
1181                 if (!ip_route_output_key(&rt, &fl)) {
1182                         tdev = rt->u.dst.dev;
1183                         ip_rt_put(rt);
1184                 }
1185
1186                 dev->flags |= IFF_POINTOPOINT;
1187
1188 #ifdef CONFIG_NET_IPGRE_BROADCAST
1189                 if (MULTICAST(iph->daddr)) {
1190                         if (!iph->saddr)
1191                                 return -EINVAL;
1192                         dev->flags = IFF_BROADCAST;
1193                         dev->hard_header = ipgre_header;
1194                         dev->open = ipgre_open;
1195                         dev->stop = ipgre_close;
1196                 }
1197 #endif
1198         }
1199
1200         if (!tdev && tunnel->parms.link)
1201                 tdev = __dev_get_by_index(tunnel->parms.link);
1202
1203         if (tdev) {
1204                 hlen = tdev->hard_header_len;
1205                 mtu = tdev->mtu;
1206         }
1207         dev->iflink = tunnel->parms.link;
1208
1209         /* Precalculate GRE options length */
1210         if (tunnel->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
1211                 if (tunnel->parms.o_flags&GRE_CSUM)
1212                         addend += 4;
1213                 if (tunnel->parms.o_flags&GRE_KEY)
1214                         addend += 4;
1215                 if (tunnel->parms.o_flags&GRE_SEQ)
1216                         addend += 4;
1217         }
1218         dev->hard_header_len = hlen + addend;
1219         dev->mtu = mtu - addend;
1220         tunnel->hlen = addend;
1221         return 0;
1222 }
1223
1224 static int __init ipgre_fb_tunnel_init(struct net_device *dev)
1225 {
1226         struct ip_tunnel *tunnel = netdev_priv(dev);
1227         struct iphdr *iph = &tunnel->parms.iph;
1228
1229         tunnel->dev = dev;
1230         strcpy(tunnel->parms.name, dev->name);
1231
1232         iph->version            = 4;
1233         iph->protocol           = IPPROTO_GRE;
1234         iph->ihl                = 5;
1235         tunnel->hlen            = sizeof(struct iphdr) + 4;
1236
1237         dev_hold(dev);
1238         tunnels_wc[0]           = tunnel;
1239         return 0;
1240 }
1241
1242
1243 static struct net_protocol ipgre_protocol = {
1244         .handler        =       ipgre_rcv,
1245         .err_handler    =       ipgre_err,
1246 };
1247
1248
1249 /*
1250  *      And now the modules code and kernel interface.
1251  */
1252
1253 static int __init ipgre_init(void)
1254 {
1255         int err;
1256
1257         printk(KERN_INFO "GRE over IPv4 tunneling driver\n");
1258
1259         if (inet_add_protocol(&ipgre_protocol, IPPROTO_GRE) < 0) {
1260                 printk(KERN_INFO "ipgre init: can't add protocol\n");
1261                 return -EAGAIN;
1262         }
1263
1264         ipgre_fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "gre0",
1265                                            ipgre_tunnel_setup);
1266         if (!ipgre_fb_tunnel_dev) {
1267                 err = -ENOMEM;
1268                 goto err1;
1269         }
1270
1271         ipgre_fb_tunnel_dev->init = ipgre_fb_tunnel_init;
1272
1273         if ((err = register_netdev(ipgre_fb_tunnel_dev)))
1274                 goto err2;
1275 out:
1276         return err;
1277 err2:
1278         free_netdev(ipgre_fb_tunnel_dev);
1279 err1:
1280         inet_del_protocol(&ipgre_protocol, IPPROTO_GRE);
1281         goto out;
1282 }
1283
1284 static void __exit ipgre_destroy_tunnels(void)
1285 {
1286         int prio;
1287
1288         for (prio = 0; prio < 4; prio++) {
1289                 int h;
1290                 for (h = 0; h < HASH_SIZE; h++) {
1291                         struct ip_tunnel *t;
1292                         while ((t = tunnels[prio][h]) != NULL)
1293                                 unregister_netdevice(t->dev);
1294                 }
1295         }
1296 }
1297
1298 static void __exit ipgre_fini(void)
1299 {
1300         if (inet_del_protocol(&ipgre_protocol, IPPROTO_GRE) < 0)
1301                 printk(KERN_INFO "ipgre close: can't remove protocol\n");
1302
1303         rtnl_lock();
1304         ipgre_destroy_tunnels();
1305         rtnl_unlock();
1306 }
1307
1308 module_init(ipgre_init);
1309 module_exit(ipgre_fini);
1310 MODULE_LICENSE("GPL");