[NETNS]: Add namespace parameter to ip_route_output_key.
[linux-2.6.git] / net / ipv6 / sit.c
1 /*
2  *      IPv6 over IPv4 tunnel device - Simple Internet Transition (SIT)
3  *      Linux INET6 implementation
4  *
5  *      Authors:
6  *      Pedro Roque             <roque@di.fc.ul.pt>
7  *      Alexey Kuznetsov        <kuznet@ms2.inr.ac.ru>
8  *
9  *      $Id: sit.c,v 1.53 2001/09/25 05:09:53 davem Exp $
10  *
11  *      This program is free software; you can redistribute it and/or
12  *      modify it under the terms of the GNU General Public License
13  *      as published by the Free Software Foundation; either version
14  *      2 of the License, or (at your option) any later version.
15  *
16  *      Changes:
17  * Roger Venning <r.venning@telstra.com>:       6to4 support
18  * Nate Thompson <nate@thebog.net>:             6to4 support
19  * Fred L. Templin <fltemplin@acm.org>:         isatap support
20  */
21
22 #include <linux/module.h>
23 #include <linux/capability.h>
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/socket.h>
27 #include <linux/sockios.h>
28 #include <linux/net.h>
29 #include <linux/in6.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/icmp.h>
33 #include <asm/uaccess.h>
34 #include <linux/init.h>
35 #include <linux/netfilter_ipv4.h>
36 #include <linux/if_ether.h>
37
38 #include <net/sock.h>
39 #include <net/snmp.h>
40
41 #include <net/ipv6.h>
42 #include <net/protocol.h>
43 #include <net/transp_v6.h>
44 #include <net/ip6_fib.h>
45 #include <net/ip6_route.h>
46 #include <net/ndisc.h>
47 #include <net/addrconf.h>
48 #include <net/ip.h>
49 #include <net/udp.h>
50 #include <net/icmp.h>
51 #include <net/ipip.h>
52 #include <net/inet_ecn.h>
53 #include <net/xfrm.h>
54 #include <net/dsfield.h>
55
56 /*
57    This version of net/ipv6/sit.c is cloned of net/ipv4/ip_gre.c
58
59    For comments look at net/ipv4/ip_gre.c --ANK
60  */
61
62 #define HASH_SIZE  16
63 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
64
65 static int ipip6_fb_tunnel_init(struct net_device *dev);
66 static int ipip6_tunnel_init(struct net_device *dev);
67 static void ipip6_tunnel_setup(struct net_device *dev);
68
69 static struct net_device *ipip6_fb_tunnel_dev;
70
71 static struct ip_tunnel *tunnels_r_l[HASH_SIZE];
72 static struct ip_tunnel *tunnels_r[HASH_SIZE];
73 static struct ip_tunnel *tunnels_l[HASH_SIZE];
74 static struct ip_tunnel *tunnels_wc[1];
75 static struct ip_tunnel **tunnels[4] = { tunnels_wc, tunnels_l, tunnels_r, tunnels_r_l };
76
77 static DEFINE_RWLOCK(ipip6_lock);
78
79 static struct ip_tunnel * ipip6_tunnel_lookup(__be32 remote, __be32 local)
80 {
81         unsigned h0 = HASH(remote);
82         unsigned h1 = HASH(local);
83         struct ip_tunnel *t;
84
85         for (t = tunnels_r_l[h0^h1]; t; t = t->next) {
86                 if (local == t->parms.iph.saddr &&
87                     remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
88                         return t;
89         }
90         for (t = tunnels_r[h0]; t; t = t->next) {
91                 if (remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
92                         return t;
93         }
94         for (t = tunnels_l[h1]; t; t = t->next) {
95                 if (local == t->parms.iph.saddr && (t->dev->flags&IFF_UP))
96                         return t;
97         }
98         if ((t = tunnels_wc[0]) != NULL && (t->dev->flags&IFF_UP))
99                 return t;
100         return NULL;
101 }
102
103 static struct ip_tunnel **__ipip6_bucket(struct ip_tunnel_parm *parms)
104 {
105         __be32 remote = parms->iph.daddr;
106         __be32 local = parms->iph.saddr;
107         unsigned h = 0;
108         int prio = 0;
109
110         if (remote) {
111                 prio |= 2;
112                 h ^= HASH(remote);
113         }
114         if (local) {
115                 prio |= 1;
116                 h ^= HASH(local);
117         }
118         return &tunnels[prio][h];
119 }
120
121 static inline struct ip_tunnel **ipip6_bucket(struct ip_tunnel *t)
122 {
123         return __ipip6_bucket(&t->parms);
124 }
125
126 static void ipip6_tunnel_unlink(struct ip_tunnel *t)
127 {
128         struct ip_tunnel **tp;
129
130         for (tp = ipip6_bucket(t); *tp; tp = &(*tp)->next) {
131                 if (t == *tp) {
132                         write_lock_bh(&ipip6_lock);
133                         *tp = t->next;
134                         write_unlock_bh(&ipip6_lock);
135                         break;
136                 }
137         }
138 }
139
140 static void ipip6_tunnel_link(struct ip_tunnel *t)
141 {
142         struct ip_tunnel **tp = ipip6_bucket(t);
143
144         t->next = *tp;
145         write_lock_bh(&ipip6_lock);
146         *tp = t;
147         write_unlock_bh(&ipip6_lock);
148 }
149
150 static struct ip_tunnel * ipip6_tunnel_locate(struct ip_tunnel_parm *parms, int create)
151 {
152         __be32 remote = parms->iph.daddr;
153         __be32 local = parms->iph.saddr;
154         struct ip_tunnel *t, **tp, *nt;
155         struct net_device *dev;
156         char name[IFNAMSIZ];
157
158         for (tp = __ipip6_bucket(parms); (t = *tp) != NULL; tp = &t->next) {
159                 if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr)
160                         return t;
161         }
162         if (!create)
163                 goto failed;
164
165         if (parms->name[0])
166                 strlcpy(name, parms->name, IFNAMSIZ);
167         else {
168                 int i;
169                 for (i=1; i<100; i++) {
170                         sprintf(name, "sit%d", i);
171                         if (__dev_get_by_name(&init_net, name) == NULL)
172                                 break;
173                 }
174                 if (i==100)
175                         goto failed;
176         }
177
178         dev = alloc_netdev(sizeof(*t), name, ipip6_tunnel_setup);
179         if (dev == NULL)
180                 return NULL;
181
182         nt = netdev_priv(dev);
183         dev->init = ipip6_tunnel_init;
184         nt->parms = *parms;
185
186         if (parms->i_flags & SIT_ISATAP)
187                 dev->priv_flags |= IFF_ISATAP;
188
189         if (register_netdevice(dev) < 0) {
190                 free_netdev(dev);
191                 goto failed;
192         }
193
194         dev_hold(dev);
195
196         ipip6_tunnel_link(nt);
197         return nt;
198
199 failed:
200         return NULL;
201 }
202
203 static void ipip6_tunnel_uninit(struct net_device *dev)
204 {
205         if (dev == ipip6_fb_tunnel_dev) {
206                 write_lock_bh(&ipip6_lock);
207                 tunnels_wc[0] = NULL;
208                 write_unlock_bh(&ipip6_lock);
209                 dev_put(dev);
210         } else {
211                 ipip6_tunnel_unlink(netdev_priv(dev));
212                 dev_put(dev);
213         }
214 }
215
216
217 static int ipip6_err(struct sk_buff *skb, u32 info)
218 {
219 #ifndef I_WISH_WORLD_WERE_PERFECT
220
221 /* It is not :-( All the routers (except for Linux) return only
222    8 bytes of packet payload. It means, that precise relaying of
223    ICMP in the real Internet is absolutely infeasible.
224  */
225         struct iphdr *iph = (struct iphdr*)skb->data;
226         const int type = icmp_hdr(skb)->type;
227         const int code = icmp_hdr(skb)->code;
228         struct ip_tunnel *t;
229         int err;
230
231         switch (type) {
232         default:
233         case ICMP_PARAMETERPROB:
234                 return 0;
235
236         case ICMP_DEST_UNREACH:
237                 switch (code) {
238                 case ICMP_SR_FAILED:
239                 case ICMP_PORT_UNREACH:
240                         /* Impossible event. */
241                         return 0;
242                 case ICMP_FRAG_NEEDED:
243                         /* Soft state for pmtu is maintained by IP core. */
244                         return 0;
245                 default:
246                         /* All others are translated to HOST_UNREACH.
247                            rfc2003 contains "deep thoughts" about NET_UNREACH,
248                            I believe they are just ether pollution. --ANK
249                          */
250                         break;
251                 }
252                 break;
253         case ICMP_TIME_EXCEEDED:
254                 if (code != ICMP_EXC_TTL)
255                         return 0;
256                 break;
257         }
258
259         err = -ENOENT;
260
261         read_lock(&ipip6_lock);
262         t = ipip6_tunnel_lookup(iph->daddr, iph->saddr);
263         if (t == NULL || t->parms.iph.daddr == 0)
264                 goto out;
265
266         err = 0;
267         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
268                 goto out;
269
270         if (jiffies - t->err_time < IPTUNNEL_ERR_TIMEO)
271                 t->err_count++;
272         else
273                 t->err_count = 1;
274         t->err_time = jiffies;
275 out:
276         read_unlock(&ipip6_lock);
277         return err;
278 #else
279         struct iphdr *iph = (struct iphdr*)dp;
280         int hlen = iph->ihl<<2;
281         struct ipv6hdr *iph6;
282         const int type = icmp_hdr(skb)->type;
283         const int code = icmp_hdr(skb)->code;
284         int rel_type = 0;
285         int rel_code = 0;
286         int rel_info = 0;
287         struct sk_buff *skb2;
288         struct rt6_info *rt6i;
289
290         if (len < hlen + sizeof(struct ipv6hdr))
291                 return;
292         iph6 = (struct ipv6hdr*)(dp + hlen);
293
294         switch (type) {
295         default:
296                 return;
297         case ICMP_PARAMETERPROB:
298                 if (icmp_hdr(skb)->un.gateway < hlen)
299                         return;
300
301                 /* So... This guy found something strange INSIDE encapsulated
302                    packet. Well, he is fool, but what can we do ?
303                  */
304                 rel_type = ICMPV6_PARAMPROB;
305                 rel_info = icmp_hdr(skb)->un.gateway - hlen;
306                 break;
307
308         case ICMP_DEST_UNREACH:
309                 switch (code) {
310                 case ICMP_SR_FAILED:
311                 case ICMP_PORT_UNREACH:
312                         /* Impossible event. */
313                         return;
314                 case ICMP_FRAG_NEEDED:
315                         /* Too complicated case ... */
316                         return;
317                 default:
318                         /* All others are translated to HOST_UNREACH.
319                            rfc2003 contains "deep thoughts" about NET_UNREACH,
320                            I believe, it is just ether pollution. --ANK
321                          */
322                         rel_type = ICMPV6_DEST_UNREACH;
323                         rel_code = ICMPV6_ADDR_UNREACH;
324                         break;
325                 }
326                 break;
327         case ICMP_TIME_EXCEEDED:
328                 if (code != ICMP_EXC_TTL)
329                         return;
330                 rel_type = ICMPV6_TIME_EXCEED;
331                 rel_code = ICMPV6_EXC_HOPLIMIT;
332                 break;
333         }
334
335         /* Prepare fake skb to feed it to icmpv6_send */
336         skb2 = skb_clone(skb, GFP_ATOMIC);
337         if (skb2 == NULL)
338                 return 0;
339         dst_release(skb2->dst);
340         skb2->dst = NULL;
341         skb_pull(skb2, skb->data - (u8*)iph6);
342         skb_reset_network_header(skb2);
343
344         /* Try to guess incoming interface */
345         rt6i = rt6_lookup(&iph6->saddr, NULL, NULL, 0);
346         if (rt6i && rt6i->rt6i_dev) {
347                 skb2->dev = rt6i->rt6i_dev;
348
349                 rt6i = rt6_lookup(&iph6->daddr, &iph6->saddr, NULL, 0);
350
351                 if (rt6i && rt6i->rt6i_dev && rt6i->rt6i_dev->type == ARPHRD_SIT) {
352                         struct ip_tunnel *t = netdev_priv(rt6i->rt6i_dev);
353                         if (rel_type == ICMPV6_TIME_EXCEED && t->parms.iph.ttl) {
354                                 rel_type = ICMPV6_DEST_UNREACH;
355                                 rel_code = ICMPV6_ADDR_UNREACH;
356                         }
357                         icmpv6_send(skb2, rel_type, rel_code, rel_info, skb2->dev);
358                 }
359         }
360         kfree_skb(skb2);
361         return 0;
362 #endif
363 }
364
365 static inline void ipip6_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb)
366 {
367         if (INET_ECN_is_ce(iph->tos))
368                 IP6_ECN_set_ce(ipv6_hdr(skb));
369 }
370
371 /* ISATAP (RFC4214) - check source address */
372 static int
373 isatap_srcok(struct sk_buff *skb, struct iphdr *iph, struct net_device *dev)
374 {
375         struct neighbour *neigh;
376         struct dst_entry *dst;
377         struct rt6_info *rt;
378         struct flowi fl;
379         struct in6_addr *addr6;
380         struct in6_addr rtr;
381         struct ipv6hdr *iph6;
382         int ok = 0;
383
384         /* from onlink default router */
385         ipv6_addr_set(&rtr,  htonl(0xFE800000), 0, 0, 0);
386         ipv6_isatap_eui64(rtr.s6_addr + 8, iph->saddr);
387         if ((rt = rt6_get_dflt_router(&rtr, dev))) {
388                 dst_release(&rt->u.dst);
389                 return 1;
390         }
391
392         iph6 = ipv6_hdr(skb);
393         memset(&fl, 0, sizeof(fl));
394         fl.proto = iph6->nexthdr;
395         ipv6_addr_copy(&fl.fl6_dst, &iph6->saddr);
396         fl.oif = dev->ifindex;
397         security_skb_classify_flow(skb, &fl);
398
399         dst = ip6_route_output(NULL, &fl);
400         if (!dst->error && (dst->dev == dev) && (neigh = dst->neighbour)) {
401
402                 addr6 = (struct in6_addr*)&neigh->primary_key;
403
404                 /* from correct previous hop */
405                 if (ipv6_addr_is_isatap(addr6) &&
406                     (addr6->s6_addr32[3] == iph->saddr))
407                         ok = 1;
408         }
409         dst_release(dst);
410         return ok;
411 }
412
413 static int ipip6_rcv(struct sk_buff *skb)
414 {
415         struct iphdr *iph;
416         struct ip_tunnel *tunnel;
417
418         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
419                 goto out;
420
421         iph = ip_hdr(skb);
422
423         read_lock(&ipip6_lock);
424         if ((tunnel = ipip6_tunnel_lookup(iph->saddr, iph->daddr)) != NULL) {
425                 secpath_reset(skb);
426                 skb->mac_header = skb->network_header;
427                 skb_reset_network_header(skb);
428                 IPCB(skb)->flags = 0;
429                 skb->protocol = htons(ETH_P_IPV6);
430                 skb->pkt_type = PACKET_HOST;
431
432                 if ((tunnel->dev->priv_flags & IFF_ISATAP) &&
433                     !isatap_srcok(skb, iph, tunnel->dev)) {
434                         tunnel->stat.rx_errors++;
435                         read_unlock(&ipip6_lock);
436                         kfree_skb(skb);
437                         return 0;
438                 }
439                 tunnel->stat.rx_packets++;
440                 tunnel->stat.rx_bytes += skb->len;
441                 skb->dev = tunnel->dev;
442                 dst_release(skb->dst);
443                 skb->dst = NULL;
444                 nf_reset(skb);
445                 ipip6_ecn_decapsulate(iph, skb);
446                 netif_rx(skb);
447                 read_unlock(&ipip6_lock);
448                 return 0;
449         }
450
451         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
452         kfree_skb(skb);
453         read_unlock(&ipip6_lock);
454 out:
455         return 0;
456 }
457
458 /* Returns the embedded IPv4 address if the IPv6 address
459    comes from 6to4 (RFC 3056) addr space */
460
461 static inline __be32 try_6to4(struct in6_addr *v6dst)
462 {
463         __be32 dst = 0;
464
465         if (v6dst->s6_addr16[0] == htons(0x2002)) {
466                 /* 6to4 v6 addr has 16 bits prefix, 32 v4addr, 16 SLA, ... */
467                 memcpy(&dst, &v6dst->s6_addr16[1], 4);
468         }
469         return dst;
470 }
471
472 /*
473  *      This function assumes it is being called from dev_queue_xmit()
474  *      and that skb is filled properly by that function.
475  */
476
477 static int ipip6_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
478 {
479         struct ip_tunnel *tunnel = netdev_priv(dev);
480         struct net_device_stats *stats = &tunnel->stat;
481         struct iphdr  *tiph = &tunnel->parms.iph;
482         struct ipv6hdr *iph6 = ipv6_hdr(skb);
483         u8     tos = tunnel->parms.iph.tos;
484         struct rtable *rt;                      /* Route to the other host */
485         struct net_device *tdev;                        /* Device to other host */
486         struct iphdr  *iph;                     /* Our new IP header */
487         unsigned int max_headroom;              /* The extra header space needed */
488         __be32 dst = tiph->daddr;
489         int    mtu;
490         struct in6_addr *addr6;
491         int addr_type;
492
493         if (tunnel->recursion++) {
494                 tunnel->stat.collisions++;
495                 goto tx_error;
496         }
497
498         if (skb->protocol != htons(ETH_P_IPV6))
499                 goto tx_error;
500
501         /* ISATAP (RFC4214) - must come before 6to4 */
502         if (dev->priv_flags & IFF_ISATAP) {
503                 struct neighbour *neigh = NULL;
504
505                 if (skb->dst)
506                         neigh = skb->dst->neighbour;
507
508                 if (neigh == NULL) {
509                         if (net_ratelimit())
510                                 printk(KERN_DEBUG "sit: nexthop == NULL\n");
511                         goto tx_error;
512                 }
513
514                 addr6 = (struct in6_addr*)&neigh->primary_key;
515                 addr_type = ipv6_addr_type(addr6);
516
517                 if ((addr_type & IPV6_ADDR_UNICAST) &&
518                      ipv6_addr_is_isatap(addr6))
519                         dst = addr6->s6_addr32[3];
520                 else
521                         goto tx_error;
522         }
523
524         if (!dst)
525                 dst = try_6to4(&iph6->daddr);
526
527         if (!dst) {
528                 struct neighbour *neigh = NULL;
529
530                 if (skb->dst)
531                         neigh = skb->dst->neighbour;
532
533                 if (neigh == NULL) {
534                         if (net_ratelimit())
535                                 printk(KERN_DEBUG "sit: nexthop == NULL\n");
536                         goto tx_error;
537                 }
538
539                 addr6 = (struct in6_addr*)&neigh->primary_key;
540                 addr_type = ipv6_addr_type(addr6);
541
542                 if (addr_type == IPV6_ADDR_ANY) {
543                         addr6 = &ipv6_hdr(skb)->daddr;
544                         addr_type = ipv6_addr_type(addr6);
545                 }
546
547                 if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
548                         goto tx_error_icmp;
549
550                 dst = addr6->s6_addr32[3];
551         }
552
553         {
554                 struct flowi fl = { .nl_u = { .ip4_u =
555                                               { .daddr = dst,
556                                                 .saddr = tiph->saddr,
557                                                 .tos = RT_TOS(tos) } },
558                                     .oif = tunnel->parms.link,
559                                     .proto = IPPROTO_IPV6 };
560                 if (ip_route_output_key(&init_net, &rt, &fl)) {
561                         tunnel->stat.tx_carrier_errors++;
562                         goto tx_error_icmp;
563                 }
564         }
565         if (rt->rt_type != RTN_UNICAST) {
566                 ip_rt_put(rt);
567                 tunnel->stat.tx_carrier_errors++;
568                 goto tx_error_icmp;
569         }
570         tdev = rt->u.dst.dev;
571
572         if (tdev == dev) {
573                 ip_rt_put(rt);
574                 tunnel->stat.collisions++;
575                 goto tx_error;
576         }
577
578         if (tiph->frag_off)
579                 mtu = dst_mtu(&rt->u.dst) - sizeof(struct iphdr);
580         else
581                 mtu = skb->dst ? dst_mtu(skb->dst) : dev->mtu;
582
583         if (mtu < 68) {
584                 tunnel->stat.collisions++;
585                 ip_rt_put(rt);
586                 goto tx_error;
587         }
588         if (mtu < IPV6_MIN_MTU)
589                 mtu = IPV6_MIN_MTU;
590         if (tunnel->parms.iph.daddr && skb->dst)
591                 skb->dst->ops->update_pmtu(skb->dst, mtu);
592
593         if (skb->len > mtu) {
594                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, dev);
595                 ip_rt_put(rt);
596                 goto tx_error;
597         }
598
599         if (tunnel->err_count > 0) {
600                 if (jiffies - tunnel->err_time < IPTUNNEL_ERR_TIMEO) {
601                         tunnel->err_count--;
602                         dst_link_failure(skb);
603                 } else
604                         tunnel->err_count = 0;
605         }
606
607         /*
608          * Okay, now see if we can stuff it in the buffer as-is.
609          */
610         max_headroom = LL_RESERVED_SPACE(tdev)+sizeof(struct iphdr);
611
612         if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
613             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
614                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
615                 if (!new_skb) {
616                         ip_rt_put(rt);
617                         stats->tx_dropped++;
618                         dev_kfree_skb(skb);
619                         tunnel->recursion--;
620                         return 0;
621                 }
622                 if (skb->sk)
623                         skb_set_owner_w(new_skb, skb->sk);
624                 dev_kfree_skb(skb);
625                 skb = new_skb;
626                 iph6 = ipv6_hdr(skb);
627         }
628
629         skb->transport_header = skb->network_header;
630         skb_push(skb, sizeof(struct iphdr));
631         skb_reset_network_header(skb);
632         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
633         IPCB(skb)->flags = 0;
634         dst_release(skb->dst);
635         skb->dst = &rt->u.dst;
636
637         /*
638          *      Push down and install the IPIP header.
639          */
640
641         iph                     =       ip_hdr(skb);
642         iph->version            =       4;
643         iph->ihl                =       sizeof(struct iphdr)>>2;
644         if (mtu > IPV6_MIN_MTU)
645                 iph->frag_off   =       htons(IP_DF);
646         else
647                 iph->frag_off   =       0;
648
649         iph->protocol           =       IPPROTO_IPV6;
650         iph->tos                =       INET_ECN_encapsulate(tos, ipv6_get_dsfield(iph6));
651         iph->daddr              =       rt->rt_dst;
652         iph->saddr              =       rt->rt_src;
653
654         if ((iph->ttl = tiph->ttl) == 0)
655                 iph->ttl        =       iph6->hop_limit;
656
657         nf_reset(skb);
658
659         IPTUNNEL_XMIT();
660         tunnel->recursion--;
661         return 0;
662
663 tx_error_icmp:
664         dst_link_failure(skb);
665 tx_error:
666         stats->tx_errors++;
667         dev_kfree_skb(skb);
668         tunnel->recursion--;
669         return 0;
670 }
671
672 static void ipip6_tunnel_bind_dev(struct net_device *dev)
673 {
674         struct net_device *tdev = NULL;
675         struct ip_tunnel *tunnel;
676         struct iphdr *iph;
677
678         tunnel = netdev_priv(dev);
679         iph = &tunnel->parms.iph;
680
681         if (iph->daddr) {
682                 struct flowi fl = { .nl_u = { .ip4_u =
683                                               { .daddr = iph->daddr,
684                                                 .saddr = iph->saddr,
685                                                 .tos = RT_TOS(iph->tos) } },
686                                     .oif = tunnel->parms.link,
687                                     .proto = IPPROTO_IPV6 };
688                 struct rtable *rt;
689                 if (!ip_route_output_key(&init_net, &rt, &fl)) {
690                         tdev = rt->u.dst.dev;
691                         ip_rt_put(rt);
692                 }
693                 dev->flags |= IFF_POINTOPOINT;
694         }
695
696         if (!tdev && tunnel->parms.link)
697                 tdev = __dev_get_by_index(&init_net, tunnel->parms.link);
698
699         if (tdev) {
700                 dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
701                 dev->mtu = tdev->mtu - sizeof(struct iphdr);
702                 if (dev->mtu < IPV6_MIN_MTU)
703                         dev->mtu = IPV6_MIN_MTU;
704         }
705         dev->iflink = tunnel->parms.link;
706 }
707
708 static int
709 ipip6_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
710 {
711         int err = 0;
712         struct ip_tunnel_parm p;
713         struct ip_tunnel *t;
714
715         switch (cmd) {
716         case SIOCGETTUNNEL:
717                 t = NULL;
718                 if (dev == ipip6_fb_tunnel_dev) {
719                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
720                                 err = -EFAULT;
721                                 break;
722                         }
723                         t = ipip6_tunnel_locate(&p, 0);
724                 }
725                 if (t == NULL)
726                         t = netdev_priv(dev);
727                 memcpy(&p, &t->parms, sizeof(p));
728                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
729                         err = -EFAULT;
730                 break;
731
732         case SIOCADDTUNNEL:
733         case SIOCCHGTUNNEL:
734                 err = -EPERM;
735                 if (!capable(CAP_NET_ADMIN))
736                         goto done;
737
738                 err = -EFAULT;
739                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
740                         goto done;
741
742                 err = -EINVAL;
743                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPV6 ||
744                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
745                         goto done;
746                 if (p.iph.ttl)
747                         p.iph.frag_off |= htons(IP_DF);
748
749                 t = ipip6_tunnel_locate(&p, cmd == SIOCADDTUNNEL);
750
751                 if (dev != ipip6_fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
752                         if (t != NULL) {
753                                 if (t->dev != dev) {
754                                         err = -EEXIST;
755                                         break;
756                                 }
757                         } else {
758                                 if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
759                                     (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
760                                         err = -EINVAL;
761                                         break;
762                                 }
763                                 t = netdev_priv(dev);
764                                 ipip6_tunnel_unlink(t);
765                                 t->parms.iph.saddr = p.iph.saddr;
766                                 t->parms.iph.daddr = p.iph.daddr;
767                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
768                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
769                                 ipip6_tunnel_link(t);
770                                 netdev_state_change(dev);
771                         }
772                 }
773
774                 if (t) {
775                         err = 0;
776                         if (cmd == SIOCCHGTUNNEL) {
777                                 t->parms.iph.ttl = p.iph.ttl;
778                                 t->parms.iph.tos = p.iph.tos;
779                                 if (t->parms.link != p.link) {
780                                         t->parms.link = p.link;
781                                         ipip6_tunnel_bind_dev(dev);
782                                         netdev_state_change(dev);
783                                 }
784                         }
785                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
786                                 err = -EFAULT;
787                 } else
788                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
789                 break;
790
791         case SIOCDELTUNNEL:
792                 err = -EPERM;
793                 if (!capable(CAP_NET_ADMIN))
794                         goto done;
795
796                 if (dev == ipip6_fb_tunnel_dev) {
797                         err = -EFAULT;
798                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
799                                 goto done;
800                         err = -ENOENT;
801                         if ((t = ipip6_tunnel_locate(&p, 0)) == NULL)
802                                 goto done;
803                         err = -EPERM;
804                         if (t == netdev_priv(ipip6_fb_tunnel_dev))
805                                 goto done;
806                         dev = t->dev;
807                 }
808                 unregister_netdevice(dev);
809                 err = 0;
810                 break;
811
812         default:
813                 err = -EINVAL;
814         }
815
816 done:
817         return err;
818 }
819
820 static struct net_device_stats *ipip6_tunnel_get_stats(struct net_device *dev)
821 {
822         return &(((struct ip_tunnel*)netdev_priv(dev))->stat);
823 }
824
825 static int ipip6_tunnel_change_mtu(struct net_device *dev, int new_mtu)
826 {
827         if (new_mtu < IPV6_MIN_MTU || new_mtu > 0xFFF8 - sizeof(struct iphdr))
828                 return -EINVAL;
829         dev->mtu = new_mtu;
830         return 0;
831 }
832
833 static void ipip6_tunnel_setup(struct net_device *dev)
834 {
835         dev->uninit             = ipip6_tunnel_uninit;
836         dev->destructor         = free_netdev;
837         dev->hard_start_xmit    = ipip6_tunnel_xmit;
838         dev->get_stats          = ipip6_tunnel_get_stats;
839         dev->do_ioctl           = ipip6_tunnel_ioctl;
840         dev->change_mtu         = ipip6_tunnel_change_mtu;
841
842         dev->type               = ARPHRD_SIT;
843         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr);
844         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr);
845         dev->flags              = IFF_NOARP;
846         dev->iflink             = 0;
847         dev->addr_len           = 4;
848 }
849
850 static int ipip6_tunnel_init(struct net_device *dev)
851 {
852         struct ip_tunnel *tunnel;
853
854         tunnel = netdev_priv(dev);
855
856         tunnel->dev = dev;
857         strcpy(tunnel->parms.name, dev->name);
858
859         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
860         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
861
862         ipip6_tunnel_bind_dev(dev);
863
864         return 0;
865 }
866
867 static int __init ipip6_fb_tunnel_init(struct net_device *dev)
868 {
869         struct ip_tunnel *tunnel = netdev_priv(dev);
870         struct iphdr *iph = &tunnel->parms.iph;
871
872         tunnel->dev = dev;
873         strcpy(tunnel->parms.name, dev->name);
874
875         iph->version            = 4;
876         iph->protocol           = IPPROTO_IPV6;
877         iph->ihl                = 5;
878         iph->ttl                = 64;
879
880         dev_hold(dev);
881         tunnels_wc[0]           = tunnel;
882         return 0;
883 }
884
885 static struct xfrm_tunnel sit_handler = {
886         .handler        =       ipip6_rcv,
887         .err_handler    =       ipip6_err,
888         .priority       =       1,
889 };
890
891 static void __exit sit_destroy_tunnels(void)
892 {
893         int prio;
894
895         for (prio = 1; prio < 4; prio++) {
896                 int h;
897                 for (h = 0; h < HASH_SIZE; h++) {
898                         struct ip_tunnel *t;
899                         while ((t = tunnels[prio][h]) != NULL)
900                                 unregister_netdevice(t->dev);
901                 }
902         }
903 }
904
905 static void __exit sit_cleanup(void)
906 {
907         xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
908
909         rtnl_lock();
910         sit_destroy_tunnels();
911         unregister_netdevice(ipip6_fb_tunnel_dev);
912         rtnl_unlock();
913 }
914
915 static int __init sit_init(void)
916 {
917         int err;
918
919         printk(KERN_INFO "IPv6 over IPv4 tunneling driver\n");
920
921         if (xfrm4_tunnel_register(&sit_handler, AF_INET6) < 0) {
922                 printk(KERN_INFO "sit init: Can't add protocol\n");
923                 return -EAGAIN;
924         }
925
926         ipip6_fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "sit0",
927                                            ipip6_tunnel_setup);
928         if (!ipip6_fb_tunnel_dev) {
929                 err = -ENOMEM;
930                 goto err1;
931         }
932
933         ipip6_fb_tunnel_dev->init = ipip6_fb_tunnel_init;
934
935         if ((err =  register_netdev(ipip6_fb_tunnel_dev)))
936                 goto err2;
937
938  out:
939         return err;
940  err2:
941         free_netdev(ipip6_fb_tunnel_dev);
942  err1:
943         xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
944         goto out;
945 }
946
947 module_init(sit_init);
948 module_exit(sit_cleanup);
949 MODULE_LICENSE("GPL");
950 MODULE_ALIAS("sit0");