4699cd3c3118ffefb90d4f1e587683aedddcc866
[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  *      This program is free software; you can redistribute it and/or
10  *      modify it under the terms of the GNU General Public License
11  *      as published by the Free Software Foundation; either version
12  *      2 of the License, or (at your option) any later version.
13  *
14  *      Changes:
15  * Roger Venning <r.venning@telstra.com>:       6to4 support
16  * Nate Thompson <nate@thebog.net>:             6to4 support
17  * Fred Templin <fred.l.templin@boeing.com>:    isatap support
18  */
19
20 #include <linux/module.h>
21 #include <linux/capability.h>
22 #include <linux/errno.h>
23 #include <linux/types.h>
24 #include <linux/socket.h>
25 #include <linux/sockios.h>
26 #include <linux/net.h>
27 #include <linux/in6.h>
28 #include <linux/netdevice.h>
29 #include <linux/if_arp.h>
30 #include <linux/icmp.h>
31 #include <linux/slab.h>
32 #include <asm/uaccess.h>
33 #include <linux/init.h>
34 #include <linux/netfilter_ipv4.h>
35 #include <linux/if_ether.h>
36
37 #include <net/sock.h>
38 #include <net/snmp.h>
39
40 #include <net/ipv6.h>
41 #include <net/protocol.h>
42 #include <net/transp_v6.h>
43 #include <net/ip6_fib.h>
44 #include <net/ip6_route.h>
45 #include <net/ndisc.h>
46 #include <net/addrconf.h>
47 #include <net/ip.h>
48 #include <net/udp.h>
49 #include <net/icmp.h>
50 #include <net/ipip.h>
51 #include <net/inet_ecn.h>
52 #include <net/xfrm.h>
53 #include <net/dsfield.h>
54 #include <net/net_namespace.h>
55 #include <net/netns/generic.h>
56
57 /*
58    This version of net/ipv6/sit.c is cloned of net/ipv4/ip_gre.c
59
60    For comments look at net/ipv4/ip_gre.c --ANK
61  */
62
63 #define HASH_SIZE  16
64 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
65
66 static void ipip6_tunnel_init(struct net_device *dev);
67 static void ipip6_tunnel_setup(struct net_device *dev);
68
69 static int sit_net_id __read_mostly;
70 struct sit_net {
71         struct ip_tunnel *tunnels_r_l[HASH_SIZE];
72         struct ip_tunnel *tunnels_r[HASH_SIZE];
73         struct ip_tunnel *tunnels_l[HASH_SIZE];
74         struct ip_tunnel *tunnels_wc[1];
75         struct ip_tunnel **tunnels[4];
76
77         struct net_device *fb_tunnel_dev;
78 };
79
80 /*
81  * Locking : hash tables are protected by RCU and a spinlock
82  */
83 static DEFINE_SPINLOCK(ipip6_lock);
84
85 #define for_each_ip_tunnel_rcu(start) \
86         for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
87
88 /*
89  * Must be invoked with rcu_read_lock
90  */
91 static struct ip_tunnel * ipip6_tunnel_lookup(struct net *net,
92                 struct net_device *dev, __be32 remote, __be32 local)
93 {
94         unsigned h0 = HASH(remote);
95         unsigned h1 = HASH(local);
96         struct ip_tunnel *t;
97         struct sit_net *sitn = net_generic(net, sit_net_id);
98
99         for_each_ip_tunnel_rcu(sitn->tunnels_r_l[h0 ^ h1]) {
100                 if (local == t->parms.iph.saddr &&
101                     remote == t->parms.iph.daddr &&
102                     (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
103                     (t->dev->flags & IFF_UP))
104                         return t;
105         }
106         for_each_ip_tunnel_rcu(sitn->tunnels_r[h0]) {
107                 if (remote == t->parms.iph.daddr &&
108                     (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
109                     (t->dev->flags & IFF_UP))
110                         return t;
111         }
112         for_each_ip_tunnel_rcu(sitn->tunnels_l[h1]) {
113                 if (local == t->parms.iph.saddr &&
114                     (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
115                     (t->dev->flags & IFF_UP))
116                         return t;
117         }
118         t = rcu_dereference(sitn->tunnels_wc[0]);
119         if ((t != NULL) && (t->dev->flags & IFF_UP))
120                 return t;
121         return NULL;
122 }
123
124 static struct ip_tunnel **__ipip6_bucket(struct sit_net *sitn,
125                 struct ip_tunnel_parm *parms)
126 {
127         __be32 remote = parms->iph.daddr;
128         __be32 local = parms->iph.saddr;
129         unsigned h = 0;
130         int prio = 0;
131
132         if (remote) {
133                 prio |= 2;
134                 h ^= HASH(remote);
135         }
136         if (local) {
137                 prio |= 1;
138                 h ^= HASH(local);
139         }
140         return &sitn->tunnels[prio][h];
141 }
142
143 static inline struct ip_tunnel **ipip6_bucket(struct sit_net *sitn,
144                 struct ip_tunnel *t)
145 {
146         return __ipip6_bucket(sitn, &t->parms);
147 }
148
149 static void ipip6_tunnel_unlink(struct sit_net *sitn, struct ip_tunnel *t)
150 {
151         struct ip_tunnel **tp;
152
153         for (tp = ipip6_bucket(sitn, t); *tp; tp = &(*tp)->next) {
154                 if (t == *tp) {
155                         spin_lock_bh(&ipip6_lock);
156                         *tp = t->next;
157                         spin_unlock_bh(&ipip6_lock);
158                         break;
159                 }
160         }
161 }
162
163 static void ipip6_tunnel_link(struct sit_net *sitn, struct ip_tunnel *t)
164 {
165         struct ip_tunnel **tp = ipip6_bucket(sitn, t);
166
167         spin_lock_bh(&ipip6_lock);
168         t->next = *tp;
169         rcu_assign_pointer(*tp, t);
170         spin_unlock_bh(&ipip6_lock);
171 }
172
173 static void ipip6_tunnel_clone_6rd(struct net_device *dev, struct sit_net *sitn)
174 {
175 #ifdef CONFIG_IPV6_SIT_6RD
176         struct ip_tunnel *t = netdev_priv(dev);
177
178         if (t->dev == sitn->fb_tunnel_dev) {
179                 ipv6_addr_set(&t->ip6rd.prefix, htonl(0x20020000), 0, 0, 0);
180                 t->ip6rd.relay_prefix = 0;
181                 t->ip6rd.prefixlen = 16;
182                 t->ip6rd.relay_prefixlen = 0;
183         } else {
184                 struct ip_tunnel *t0 = netdev_priv(sitn->fb_tunnel_dev);
185                 memcpy(&t->ip6rd, &t0->ip6rd, sizeof(t->ip6rd));
186         }
187 #endif
188 }
189
190 static struct ip_tunnel * ipip6_tunnel_locate(struct net *net,
191                 struct ip_tunnel_parm *parms, int create)
192 {
193         __be32 remote = parms->iph.daddr;
194         __be32 local = parms->iph.saddr;
195         struct ip_tunnel *t, **tp, *nt;
196         struct net_device *dev;
197         char name[IFNAMSIZ];
198         struct sit_net *sitn = net_generic(net, sit_net_id);
199
200         for (tp = __ipip6_bucket(sitn, parms); (t = *tp) != NULL; tp = &t->next) {
201                 if (local == t->parms.iph.saddr &&
202                     remote == t->parms.iph.daddr &&
203                     parms->link == t->parms.link) {
204                         if (create)
205                                 return NULL;
206                         else
207                                 return t;
208                 }
209         }
210         if (!create)
211                 goto failed;
212
213         if (parms->name[0])
214                 strlcpy(name, parms->name, IFNAMSIZ);
215         else
216                 sprintf(name, "sit%%d");
217
218         dev = alloc_netdev(sizeof(*t), name, ipip6_tunnel_setup);
219         if (dev == NULL)
220                 return NULL;
221
222         dev_net_set(dev, net);
223
224         if (strchr(name, '%')) {
225                 if (dev_alloc_name(dev, name) < 0)
226                         goto failed_free;
227         }
228
229         nt = netdev_priv(dev);
230
231         nt->parms = *parms;
232         ipip6_tunnel_init(dev);
233         ipip6_tunnel_clone_6rd(dev, sitn);
234
235         if (parms->i_flags & SIT_ISATAP)
236                 dev->priv_flags |= IFF_ISATAP;
237
238         if (register_netdevice(dev) < 0)
239                 goto failed_free;
240
241         dev_hold(dev);
242
243         ipip6_tunnel_link(sitn, nt);
244         return nt;
245
246 failed_free:
247         free_netdev(dev);
248 failed:
249         return NULL;
250 }
251
252 #define for_each_prl_rcu(start)                 \
253         for (prl = rcu_dereference(start);      \
254              prl;                               \
255              prl = rcu_dereference(prl->next))
256
257 static struct ip_tunnel_prl_entry *
258 __ipip6_tunnel_locate_prl(struct ip_tunnel *t, __be32 addr)
259 {
260         struct ip_tunnel_prl_entry *prl;
261
262         for_each_prl_rcu(t->prl)
263                 if (prl->addr == addr)
264                         break;
265         return prl;
266
267 }
268
269 static int ipip6_tunnel_get_prl(struct ip_tunnel *t,
270                                 struct ip_tunnel_prl __user *a)
271 {
272         struct ip_tunnel_prl kprl, *kp;
273         struct ip_tunnel_prl_entry *prl;
274         unsigned int cmax, c = 0, ca, len;
275         int ret = 0;
276
277         if (copy_from_user(&kprl, a, sizeof(kprl)))
278                 return -EFAULT;
279         cmax = kprl.datalen / sizeof(kprl);
280         if (cmax > 1 && kprl.addr != htonl(INADDR_ANY))
281                 cmax = 1;
282
283         /* For simple GET or for root users,
284          * we try harder to allocate.
285          */
286         kp = (cmax <= 1 || capable(CAP_NET_ADMIN)) ?
287                 kcalloc(cmax, sizeof(*kp), GFP_KERNEL) :
288                 NULL;
289
290         rcu_read_lock();
291
292         ca = t->prl_count < cmax ? t->prl_count : cmax;
293
294         if (!kp) {
295                 /* We don't try hard to allocate much memory for
296                  * non-root users.
297                  * For root users, retry allocating enough memory for
298                  * the answer.
299                  */
300                 kp = kcalloc(ca, sizeof(*kp), GFP_ATOMIC);
301                 if (!kp) {
302                         ret = -ENOMEM;
303                         goto out;
304                 }
305         }
306
307         c = 0;
308         for_each_prl_rcu(t->prl) {
309                 if (c >= cmax)
310                         break;
311                 if (kprl.addr != htonl(INADDR_ANY) && prl->addr != kprl.addr)
312                         continue;
313                 kp[c].addr = prl->addr;
314                 kp[c].flags = prl->flags;
315                 c++;
316                 if (kprl.addr != htonl(INADDR_ANY))
317                         break;
318         }
319 out:
320         rcu_read_unlock();
321
322         len = sizeof(*kp) * c;
323         ret = 0;
324         if ((len && copy_to_user(a + 1, kp, len)) || put_user(len, &a->datalen))
325                 ret = -EFAULT;
326
327         kfree(kp);
328
329         return ret;
330 }
331
332 static int
333 ipip6_tunnel_add_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a, int chg)
334 {
335         struct ip_tunnel_prl_entry *p;
336         int err = 0;
337
338         if (a->addr == htonl(INADDR_ANY))
339                 return -EINVAL;
340
341         ASSERT_RTNL();
342
343         for (p = t->prl; p; p = p->next) {
344                 if (p->addr == a->addr) {
345                         if (chg) {
346                                 p->flags = a->flags;
347                                 goto out;
348                         }
349                         err = -EEXIST;
350                         goto out;
351                 }
352         }
353
354         if (chg) {
355                 err = -ENXIO;
356                 goto out;
357         }
358
359         p = kzalloc(sizeof(struct ip_tunnel_prl_entry), GFP_KERNEL);
360         if (!p) {
361                 err = -ENOBUFS;
362                 goto out;
363         }
364
365         p->next = t->prl;
366         p->addr = a->addr;
367         p->flags = a->flags;
368         t->prl_count++;
369         rcu_assign_pointer(t->prl, p);
370 out:
371         return err;
372 }
373
374 static void prl_entry_destroy_rcu(struct rcu_head *head)
375 {
376         kfree(container_of(head, struct ip_tunnel_prl_entry, rcu_head));
377 }
378
379 static void prl_list_destroy_rcu(struct rcu_head *head)
380 {
381         struct ip_tunnel_prl_entry *p, *n;
382
383         p = container_of(head, struct ip_tunnel_prl_entry, rcu_head);
384         do {
385                 n = p->next;
386                 kfree(p);
387                 p = n;
388         } while (p);
389 }
390
391 static int
392 ipip6_tunnel_del_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a)
393 {
394         struct ip_tunnel_prl_entry *x, **p;
395         int err = 0;
396
397         ASSERT_RTNL();
398
399         if (a && a->addr != htonl(INADDR_ANY)) {
400                 for (p = &t->prl; *p; p = &(*p)->next) {
401                         if ((*p)->addr == a->addr) {
402                                 x = *p;
403                                 *p = x->next;
404                                 call_rcu(&x->rcu_head, prl_entry_destroy_rcu);
405                                 t->prl_count--;
406                                 goto out;
407                         }
408                 }
409                 err = -ENXIO;
410         } else {
411                 if (t->prl) {
412                         t->prl_count = 0;
413                         x = t->prl;
414                         call_rcu(&x->rcu_head, prl_list_destroy_rcu);
415                         t->prl = NULL;
416                 }
417         }
418 out:
419         return err;
420 }
421
422 static int
423 isatap_chksrc(struct sk_buff *skb, struct iphdr *iph, struct ip_tunnel *t)
424 {
425         struct ip_tunnel_prl_entry *p;
426         int ok = 1;
427
428         rcu_read_lock();
429         p = __ipip6_tunnel_locate_prl(t, iph->saddr);
430         if (p) {
431                 if (p->flags & PRL_DEFAULT)
432                         skb->ndisc_nodetype = NDISC_NODETYPE_DEFAULT;
433                 else
434                         skb->ndisc_nodetype = NDISC_NODETYPE_NODEFAULT;
435         } else {
436                 struct in6_addr *addr6 = &ipv6_hdr(skb)->saddr;
437                 if (ipv6_addr_is_isatap(addr6) &&
438                     (addr6->s6_addr32[3] == iph->saddr) &&
439                     ipv6_chk_prefix(addr6, t->dev))
440                         skb->ndisc_nodetype = NDISC_NODETYPE_HOST;
441                 else
442                         ok = 0;
443         }
444         rcu_read_unlock();
445         return ok;
446 }
447
448 static void ipip6_tunnel_uninit(struct net_device *dev)
449 {
450         struct net *net = dev_net(dev);
451         struct sit_net *sitn = net_generic(net, sit_net_id);
452
453         if (dev == sitn->fb_tunnel_dev) {
454                 spin_lock_bh(&ipip6_lock);
455                 sitn->tunnels_wc[0] = NULL;
456                 spin_unlock_bh(&ipip6_lock);
457                 dev_put(dev);
458         } else {
459                 ipip6_tunnel_unlink(sitn, netdev_priv(dev));
460                 ipip6_tunnel_del_prl(netdev_priv(dev), NULL);
461                 dev_put(dev);
462         }
463 }
464
465
466 static int ipip6_err(struct sk_buff *skb, u32 info)
467 {
468
469 /* All the routers (except for Linux) return only
470    8 bytes of packet payload. It means, that precise relaying of
471    ICMP in the real Internet is absolutely infeasible.
472  */
473         struct iphdr *iph = (struct iphdr*)skb->data;
474         const int type = icmp_hdr(skb)->type;
475         const int code = icmp_hdr(skb)->code;
476         struct ip_tunnel *t;
477         int err;
478
479         switch (type) {
480         default:
481         case ICMP_PARAMETERPROB:
482                 return 0;
483
484         case ICMP_DEST_UNREACH:
485                 switch (code) {
486                 case ICMP_SR_FAILED:
487                 case ICMP_PORT_UNREACH:
488                         /* Impossible event. */
489                         return 0;
490                 case ICMP_FRAG_NEEDED:
491                         /* Soft state for pmtu is maintained by IP core. */
492                         return 0;
493                 default:
494                         /* All others are translated to HOST_UNREACH.
495                            rfc2003 contains "deep thoughts" about NET_UNREACH,
496                            I believe they are just ether pollution. --ANK
497                          */
498                         break;
499                 }
500                 break;
501         case ICMP_TIME_EXCEEDED:
502                 if (code != ICMP_EXC_TTL)
503                         return 0;
504                 break;
505         }
506
507         err = -ENOENT;
508
509         rcu_read_lock();
510         t = ipip6_tunnel_lookup(dev_net(skb->dev),
511                                 skb->dev,
512                                 iph->daddr,
513                                 iph->saddr);
514         if (t == NULL || t->parms.iph.daddr == 0)
515                 goto out;
516
517         err = 0;
518         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
519                 goto out;
520
521         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
522                 t->err_count++;
523         else
524                 t->err_count = 1;
525         t->err_time = jiffies;
526 out:
527         rcu_read_unlock();
528         return err;
529 }
530
531 static inline void ipip6_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb)
532 {
533         if (INET_ECN_is_ce(iph->tos))
534                 IP6_ECN_set_ce(ipv6_hdr(skb));
535 }
536
537 static int ipip6_rcv(struct sk_buff *skb)
538 {
539         struct iphdr *iph;
540         struct ip_tunnel *tunnel;
541
542         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
543                 goto out;
544
545         iph = ip_hdr(skb);
546
547         rcu_read_lock();
548         tunnel = ipip6_tunnel_lookup(dev_net(skb->dev), skb->dev,
549                                      iph->saddr, iph->daddr);
550         if (tunnel != NULL) {
551                 secpath_reset(skb);
552                 skb->mac_header = skb->network_header;
553                 skb_reset_network_header(skb);
554                 IPCB(skb)->flags = 0;
555                 skb->protocol = htons(ETH_P_IPV6);
556                 skb->pkt_type = PACKET_HOST;
557
558                 if ((tunnel->dev->priv_flags & IFF_ISATAP) &&
559                     !isatap_chksrc(skb, iph, tunnel)) {
560                         tunnel->dev->stats.rx_errors++;
561                         rcu_read_unlock();
562                         kfree_skb(skb);
563                         return 0;
564                 }
565
566                 skb_tunnel_rx(skb, tunnel->dev);
567
568                 ipip6_ecn_decapsulate(iph, skb);
569                 netif_rx(skb);
570                 rcu_read_unlock();
571                 return 0;
572         }
573
574         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
575         rcu_read_unlock();
576 out:
577         kfree_skb(skb);
578         return 0;
579 }
580
581 /*
582  * Returns the embedded IPv4 address if the IPv6 address
583  * comes from 6rd / 6to4 (RFC 3056) addr space.
584  */
585 static inline
586 __be32 try_6rd(struct in6_addr *v6dst, struct ip_tunnel *tunnel)
587 {
588         __be32 dst = 0;
589
590 #ifdef CONFIG_IPV6_SIT_6RD
591         if (ipv6_prefix_equal(v6dst, &tunnel->ip6rd.prefix,
592                               tunnel->ip6rd.prefixlen)) {
593                 unsigned pbw0, pbi0;
594                 int pbi1;
595                 u32 d;
596
597                 pbw0 = tunnel->ip6rd.prefixlen >> 5;
598                 pbi0 = tunnel->ip6rd.prefixlen & 0x1f;
599
600                 d = (ntohl(v6dst->s6_addr32[pbw0]) << pbi0) >>
601                     tunnel->ip6rd.relay_prefixlen;
602
603                 pbi1 = pbi0 - tunnel->ip6rd.relay_prefixlen;
604                 if (pbi1 > 0)
605                         d |= ntohl(v6dst->s6_addr32[pbw0 + 1]) >>
606                              (32 - pbi1);
607
608                 dst = tunnel->ip6rd.relay_prefix | htonl(d);
609         }
610 #else
611         if (v6dst->s6_addr16[0] == htons(0x2002)) {
612                 /* 6to4 v6 addr has 16 bits prefix, 32 v4addr, 16 SLA, ... */
613                 memcpy(&dst, &v6dst->s6_addr16[1], 4);
614         }
615 #endif
616         return dst;
617 }
618
619 /*
620  *      This function assumes it is being called from dev_queue_xmit()
621  *      and that skb is filled properly by that function.
622  */
623
624 static netdev_tx_t ipip6_tunnel_xmit(struct sk_buff *skb,
625                                      struct net_device *dev)
626 {
627         struct ip_tunnel *tunnel = netdev_priv(dev);
628         struct net_device_stats *stats = &dev->stats;
629         struct netdev_queue *txq = netdev_get_tx_queue(dev, 0);
630         struct iphdr  *tiph = &tunnel->parms.iph;
631         struct ipv6hdr *iph6 = ipv6_hdr(skb);
632         u8     tos = tunnel->parms.iph.tos;
633         __be16 df = tiph->frag_off;
634         struct rtable *rt;                      /* Route to the other host */
635         struct net_device *tdev;                        /* Device to other host */
636         struct iphdr  *iph;                     /* Our new IP header */
637         unsigned int max_headroom;              /* The extra header space needed */
638         __be32 dst = tiph->daddr;
639         int    mtu;
640         struct in6_addr *addr6;
641         int addr_type;
642
643         if (skb->protocol != htons(ETH_P_IPV6))
644                 goto tx_error;
645
646         /* ISATAP (RFC4214) - must come before 6to4 */
647         if (dev->priv_flags & IFF_ISATAP) {
648                 struct neighbour *neigh = NULL;
649
650                 if (skb_dst(skb))
651                         neigh = skb_dst(skb)->neighbour;
652
653                 if (neigh == NULL) {
654                         if (net_ratelimit())
655                                 printk(KERN_DEBUG "sit: nexthop == NULL\n");
656                         goto tx_error;
657                 }
658
659                 addr6 = (struct in6_addr*)&neigh->primary_key;
660                 addr_type = ipv6_addr_type(addr6);
661
662                 if ((addr_type & IPV6_ADDR_UNICAST) &&
663                      ipv6_addr_is_isatap(addr6))
664                         dst = addr6->s6_addr32[3];
665                 else
666                         goto tx_error;
667         }
668
669         if (!dst)
670                 dst = try_6rd(&iph6->daddr, tunnel);
671
672         if (!dst) {
673                 struct neighbour *neigh = NULL;
674
675                 if (skb_dst(skb))
676                         neigh = skb_dst(skb)->neighbour;
677
678                 if (neigh == NULL) {
679                         if (net_ratelimit())
680                                 printk(KERN_DEBUG "sit: nexthop == NULL\n");
681                         goto tx_error;
682                 }
683
684                 addr6 = (struct in6_addr*)&neigh->primary_key;
685                 addr_type = ipv6_addr_type(addr6);
686
687                 if (addr_type == IPV6_ADDR_ANY) {
688                         addr6 = &ipv6_hdr(skb)->daddr;
689                         addr_type = ipv6_addr_type(addr6);
690                 }
691
692                 if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
693                         goto tx_error_icmp;
694
695                 dst = addr6->s6_addr32[3];
696         }
697
698         {
699                 struct flowi fl = { .nl_u = { .ip4_u =
700                                               { .daddr = dst,
701                                                 .saddr = tiph->saddr,
702                                                 .tos = RT_TOS(tos) } },
703                                     .oif = tunnel->parms.link,
704                                     .proto = IPPROTO_IPV6 };
705                 if (ip_route_output_key(dev_net(dev), &rt, &fl)) {
706                         stats->tx_carrier_errors++;
707                         goto tx_error_icmp;
708                 }
709         }
710         if (rt->rt_type != RTN_UNICAST) {
711                 ip_rt_put(rt);
712                 stats->tx_carrier_errors++;
713                 goto tx_error_icmp;
714         }
715         tdev = rt->dst.dev;
716
717         if (tdev == dev) {
718                 ip_rt_put(rt);
719                 stats->collisions++;
720                 goto tx_error;
721         }
722
723         if (df) {
724                 mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
725
726                 if (mtu < 68) {
727                         stats->collisions++;
728                         ip_rt_put(rt);
729                         goto tx_error;
730                 }
731
732                 if (mtu < IPV6_MIN_MTU) {
733                         mtu = IPV6_MIN_MTU;
734                         df = 0;
735                 }
736
737                 if (tunnel->parms.iph.daddr && skb_dst(skb))
738                         skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
739
740                 if (skb->len > mtu) {
741                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
742                         ip_rt_put(rt);
743                         goto tx_error;
744                 }
745         }
746
747         if (tunnel->err_count > 0) {
748                 if (time_before(jiffies,
749                                 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
750                         tunnel->err_count--;
751                         dst_link_failure(skb);
752                 } else
753                         tunnel->err_count = 0;
754         }
755
756         /*
757          * Okay, now see if we can stuff it in the buffer as-is.
758          */
759         max_headroom = LL_RESERVED_SPACE(tdev)+sizeof(struct iphdr);
760
761         if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
762             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
763                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
764                 if (!new_skb) {
765                         ip_rt_put(rt);
766                         txq->tx_dropped++;
767                         dev_kfree_skb(skb);
768                         return NETDEV_TX_OK;
769                 }
770                 if (skb->sk)
771                         skb_set_owner_w(new_skb, skb->sk);
772                 dev_kfree_skb(skb);
773                 skb = new_skb;
774                 iph6 = ipv6_hdr(skb);
775         }
776
777         skb->transport_header = skb->network_header;
778         skb_push(skb, sizeof(struct iphdr));
779         skb_reset_network_header(skb);
780         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
781         IPCB(skb)->flags = 0;
782         skb_dst_drop(skb);
783         skb_dst_set(skb, &rt->dst);
784
785         /*
786          *      Push down and install the IPIP header.
787          */
788
789         iph                     =       ip_hdr(skb);
790         iph->version            =       4;
791         iph->ihl                =       sizeof(struct iphdr)>>2;
792         iph->frag_off           =       df;
793         iph->protocol           =       IPPROTO_IPV6;
794         iph->tos                =       INET_ECN_encapsulate(tos, ipv6_get_dsfield(iph6));
795         iph->daddr              =       rt->rt_dst;
796         iph->saddr              =       rt->rt_src;
797
798         if ((iph->ttl = tiph->ttl) == 0)
799                 iph->ttl        =       iph6->hop_limit;
800
801         nf_reset(skb);
802
803         IPTUNNEL_XMIT();
804         return NETDEV_TX_OK;
805
806 tx_error_icmp:
807         dst_link_failure(skb);
808 tx_error:
809         stats->tx_errors++;
810         dev_kfree_skb(skb);
811         return NETDEV_TX_OK;
812 }
813
814 static void ipip6_tunnel_bind_dev(struct net_device *dev)
815 {
816         struct net_device *tdev = NULL;
817         struct ip_tunnel *tunnel;
818         struct iphdr *iph;
819
820         tunnel = netdev_priv(dev);
821         iph = &tunnel->parms.iph;
822
823         if (iph->daddr) {
824                 struct flowi fl = { .nl_u = { .ip4_u =
825                                               { .daddr = iph->daddr,
826                                                 .saddr = iph->saddr,
827                                                 .tos = RT_TOS(iph->tos) } },
828                                     .oif = tunnel->parms.link,
829                                     .proto = IPPROTO_IPV6 };
830                 struct rtable *rt;
831                 if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
832                         tdev = rt->dst.dev;
833                         ip_rt_put(rt);
834                 }
835                 dev->flags |= IFF_POINTOPOINT;
836         }
837
838         if (!tdev && tunnel->parms.link)
839                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
840
841         if (tdev) {
842                 dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
843                 dev->mtu = tdev->mtu - sizeof(struct iphdr);
844                 if (dev->mtu < IPV6_MIN_MTU)
845                         dev->mtu = IPV6_MIN_MTU;
846         }
847         dev->iflink = tunnel->parms.link;
848 }
849
850 static int
851 ipip6_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
852 {
853         int err = 0;
854         struct ip_tunnel_parm p;
855         struct ip_tunnel_prl prl;
856         struct ip_tunnel *t;
857         struct net *net = dev_net(dev);
858         struct sit_net *sitn = net_generic(net, sit_net_id);
859 #ifdef CONFIG_IPV6_SIT_6RD
860         struct ip_tunnel_6rd ip6rd;
861 #endif
862
863         switch (cmd) {
864         case SIOCGETTUNNEL:
865 #ifdef CONFIG_IPV6_SIT_6RD
866         case SIOCGET6RD:
867 #endif
868                 t = NULL;
869                 if (dev == sitn->fb_tunnel_dev) {
870                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
871                                 err = -EFAULT;
872                                 break;
873                         }
874                         t = ipip6_tunnel_locate(net, &p, 0);
875                 }
876                 if (t == NULL)
877                         t = netdev_priv(dev);
878
879                 err = -EFAULT;
880                 if (cmd == SIOCGETTUNNEL) {
881                         memcpy(&p, &t->parms, sizeof(p));
882                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &p,
883                                          sizeof(p)))
884                                 goto done;
885 #ifdef CONFIG_IPV6_SIT_6RD
886                 } else {
887                         ipv6_addr_copy(&ip6rd.prefix, &t->ip6rd.prefix);
888                         ip6rd.relay_prefix = t->ip6rd.relay_prefix;
889                         ip6rd.prefixlen = t->ip6rd.prefixlen;
890                         ip6rd.relay_prefixlen = t->ip6rd.relay_prefixlen;
891                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &ip6rd,
892                                          sizeof(ip6rd)))
893                                 goto done;
894 #endif
895                 }
896                 err = 0;
897                 break;
898
899         case SIOCADDTUNNEL:
900         case SIOCCHGTUNNEL:
901                 err = -EPERM;
902                 if (!capable(CAP_NET_ADMIN))
903                         goto done;
904
905                 err = -EFAULT;
906                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
907                         goto done;
908
909                 err = -EINVAL;
910                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPV6 ||
911                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
912                         goto done;
913                 if (p.iph.ttl)
914                         p.iph.frag_off |= htons(IP_DF);
915
916                 t = ipip6_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
917
918                 if (dev != sitn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
919                         if (t != NULL) {
920                                 if (t->dev != dev) {
921                                         err = -EEXIST;
922                                         break;
923                                 }
924                         } else {
925                                 if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
926                                     (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
927                                         err = -EINVAL;
928                                         break;
929                                 }
930                                 t = netdev_priv(dev);
931                                 ipip6_tunnel_unlink(sitn, t);
932                                 t->parms.iph.saddr = p.iph.saddr;
933                                 t->parms.iph.daddr = p.iph.daddr;
934                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
935                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
936                                 ipip6_tunnel_link(sitn, t);
937                                 netdev_state_change(dev);
938                         }
939                 }
940
941                 if (t) {
942                         err = 0;
943                         if (cmd == SIOCCHGTUNNEL) {
944                                 t->parms.iph.ttl = p.iph.ttl;
945                                 t->parms.iph.tos = p.iph.tos;
946                                 if (t->parms.link != p.link) {
947                                         t->parms.link = p.link;
948                                         ipip6_tunnel_bind_dev(dev);
949                                         netdev_state_change(dev);
950                                 }
951                         }
952                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
953                                 err = -EFAULT;
954                 } else
955                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
956                 break;
957
958         case SIOCDELTUNNEL:
959                 err = -EPERM;
960                 if (!capable(CAP_NET_ADMIN))
961                         goto done;
962
963                 if (dev == sitn->fb_tunnel_dev) {
964                         err = -EFAULT;
965                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
966                                 goto done;
967                         err = -ENOENT;
968                         if ((t = ipip6_tunnel_locate(net, &p, 0)) == NULL)
969                                 goto done;
970                         err = -EPERM;
971                         if (t == netdev_priv(sitn->fb_tunnel_dev))
972                                 goto done;
973                         dev = t->dev;
974                 }
975                 unregister_netdevice(dev);
976                 err = 0;
977                 break;
978
979         case SIOCGETPRL:
980                 err = -EINVAL;
981                 if (dev == sitn->fb_tunnel_dev)
982                         goto done;
983                 err = -ENOENT;
984                 if (!(t = netdev_priv(dev)))
985                         goto done;
986                 err = ipip6_tunnel_get_prl(t, ifr->ifr_ifru.ifru_data);
987                 break;
988
989         case SIOCADDPRL:
990         case SIOCDELPRL:
991         case SIOCCHGPRL:
992                 err = -EPERM;
993                 if (!capable(CAP_NET_ADMIN))
994                         goto done;
995                 err = -EINVAL;
996                 if (dev == sitn->fb_tunnel_dev)
997                         goto done;
998                 err = -EFAULT;
999                 if (copy_from_user(&prl, ifr->ifr_ifru.ifru_data, sizeof(prl)))
1000                         goto done;
1001                 err = -ENOENT;
1002                 if (!(t = netdev_priv(dev)))
1003                         goto done;
1004
1005                 switch (cmd) {
1006                 case SIOCDELPRL:
1007                         err = ipip6_tunnel_del_prl(t, &prl);
1008                         break;
1009                 case SIOCADDPRL:
1010                 case SIOCCHGPRL:
1011                         err = ipip6_tunnel_add_prl(t, &prl, cmd == SIOCCHGPRL);
1012                         break;
1013                 }
1014                 netdev_state_change(dev);
1015                 break;
1016
1017 #ifdef CONFIG_IPV6_SIT_6RD
1018         case SIOCADD6RD:
1019         case SIOCCHG6RD:
1020         case SIOCDEL6RD:
1021                 err = -EPERM;
1022                 if (!capable(CAP_NET_ADMIN))
1023                         goto done;
1024
1025                 err = -EFAULT;
1026                 if (copy_from_user(&ip6rd, ifr->ifr_ifru.ifru_data,
1027                                    sizeof(ip6rd)))
1028                         goto done;
1029
1030                 t = netdev_priv(dev);
1031
1032                 if (cmd != SIOCDEL6RD) {
1033                         struct in6_addr prefix;
1034                         __be32 relay_prefix;
1035
1036                         err = -EINVAL;
1037                         if (ip6rd.relay_prefixlen > 32 ||
1038                             ip6rd.prefixlen + (32 - ip6rd.relay_prefixlen) > 64)
1039                                 goto done;
1040
1041                         ipv6_addr_prefix(&prefix, &ip6rd.prefix,
1042                                          ip6rd.prefixlen);
1043                         if (!ipv6_addr_equal(&prefix, &ip6rd.prefix))
1044                                 goto done;
1045                         if (ip6rd.relay_prefixlen)
1046                                 relay_prefix = ip6rd.relay_prefix &
1047                                                htonl(0xffffffffUL <<
1048                                                      (32 - ip6rd.relay_prefixlen));
1049                         else
1050                                 relay_prefix = 0;
1051                         if (relay_prefix != ip6rd.relay_prefix)
1052                                 goto done;
1053
1054                         ipv6_addr_copy(&t->ip6rd.prefix, &prefix);
1055                         t->ip6rd.relay_prefix = relay_prefix;
1056                         t->ip6rd.prefixlen = ip6rd.prefixlen;
1057                         t->ip6rd.relay_prefixlen = ip6rd.relay_prefixlen;
1058                 } else
1059                         ipip6_tunnel_clone_6rd(dev, sitn);
1060
1061                 err = 0;
1062                 break;
1063 #endif
1064
1065         default:
1066                 err = -EINVAL;
1067         }
1068
1069 done:
1070         return err;
1071 }
1072
1073 static int ipip6_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1074 {
1075         if (new_mtu < IPV6_MIN_MTU || new_mtu > 0xFFF8 - sizeof(struct iphdr))
1076                 return -EINVAL;
1077         dev->mtu = new_mtu;
1078         return 0;
1079 }
1080
1081 static const struct net_device_ops ipip6_netdev_ops = {
1082         .ndo_uninit     = ipip6_tunnel_uninit,
1083         .ndo_start_xmit = ipip6_tunnel_xmit,
1084         .ndo_do_ioctl   = ipip6_tunnel_ioctl,
1085         .ndo_change_mtu = ipip6_tunnel_change_mtu,
1086 };
1087
1088 static void ipip6_tunnel_setup(struct net_device *dev)
1089 {
1090         dev->netdev_ops         = &ipip6_netdev_ops;
1091         dev->destructor         = free_netdev;
1092
1093         dev->type               = ARPHRD_SIT;
1094         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr);
1095         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr);
1096         dev->flags              = IFF_NOARP;
1097         dev->priv_flags        &= ~IFF_XMIT_DST_RELEASE;
1098         dev->iflink             = 0;
1099         dev->addr_len           = 4;
1100         dev->features           |= NETIF_F_NETNS_LOCAL;
1101 }
1102
1103 static void ipip6_tunnel_init(struct net_device *dev)
1104 {
1105         struct ip_tunnel *tunnel = netdev_priv(dev);
1106
1107         tunnel->dev = dev;
1108         strcpy(tunnel->parms.name, dev->name);
1109
1110         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1111         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1112
1113         ipip6_tunnel_bind_dev(dev);
1114 }
1115
1116 static void __net_init ipip6_fb_tunnel_init(struct net_device *dev)
1117 {
1118         struct ip_tunnel *tunnel = netdev_priv(dev);
1119         struct iphdr *iph = &tunnel->parms.iph;
1120         struct net *net = dev_net(dev);
1121         struct sit_net *sitn = net_generic(net, sit_net_id);
1122
1123         tunnel->dev = dev;
1124         strcpy(tunnel->parms.name, dev->name);
1125
1126         iph->version            = 4;
1127         iph->protocol           = IPPROTO_IPV6;
1128         iph->ihl                = 5;
1129         iph->ttl                = 64;
1130
1131         dev_hold(dev);
1132         sitn->tunnels_wc[0]     = tunnel;
1133 }
1134
1135 static struct xfrm_tunnel sit_handler = {
1136         .handler        =       ipip6_rcv,
1137         .err_handler    =       ipip6_err,
1138         .priority       =       1,
1139 };
1140
1141 static void __net_exit sit_destroy_tunnels(struct sit_net *sitn, struct list_head *head)
1142 {
1143         int prio;
1144
1145         for (prio = 1; prio < 4; prio++) {
1146                 int h;
1147                 for (h = 0; h < HASH_SIZE; h++) {
1148                         struct ip_tunnel *t = sitn->tunnels[prio][h];
1149
1150                         while (t != NULL) {
1151                                 unregister_netdevice_queue(t->dev, head);
1152                                 t = t->next;
1153                         }
1154                 }
1155         }
1156 }
1157
1158 static int __net_init sit_init_net(struct net *net)
1159 {
1160         struct sit_net *sitn = net_generic(net, sit_net_id);
1161         int err;
1162
1163         sitn->tunnels[0] = sitn->tunnels_wc;
1164         sitn->tunnels[1] = sitn->tunnels_l;
1165         sitn->tunnels[2] = sitn->tunnels_r;
1166         sitn->tunnels[3] = sitn->tunnels_r_l;
1167
1168         sitn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "sit0",
1169                                            ipip6_tunnel_setup);
1170         if (!sitn->fb_tunnel_dev) {
1171                 err = -ENOMEM;
1172                 goto err_alloc_dev;
1173         }
1174         dev_net_set(sitn->fb_tunnel_dev, net);
1175
1176         ipip6_fb_tunnel_init(sitn->fb_tunnel_dev);
1177         ipip6_tunnel_clone_6rd(sitn->fb_tunnel_dev, sitn);
1178
1179         if ((err = register_netdev(sitn->fb_tunnel_dev)))
1180                 goto err_reg_dev;
1181
1182         return 0;
1183
1184 err_reg_dev:
1185         dev_put(sitn->fb_tunnel_dev);
1186         free_netdev(sitn->fb_tunnel_dev);
1187 err_alloc_dev:
1188         return err;
1189 }
1190
1191 static void __net_exit sit_exit_net(struct net *net)
1192 {
1193         struct sit_net *sitn = net_generic(net, sit_net_id);
1194         LIST_HEAD(list);
1195
1196         rtnl_lock();
1197         sit_destroy_tunnels(sitn, &list);
1198         unregister_netdevice_queue(sitn->fb_tunnel_dev, &list);
1199         unregister_netdevice_many(&list);
1200         rtnl_unlock();
1201 }
1202
1203 static struct pernet_operations sit_net_ops = {
1204         .init = sit_init_net,
1205         .exit = sit_exit_net,
1206         .id   = &sit_net_id,
1207         .size = sizeof(struct sit_net),
1208 };
1209
1210 static void __exit sit_cleanup(void)
1211 {
1212         xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
1213
1214         unregister_pernet_device(&sit_net_ops);
1215         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1216 }
1217
1218 static int __init sit_init(void)
1219 {
1220         int err;
1221
1222         printk(KERN_INFO "IPv6 over IPv4 tunneling driver\n");
1223
1224         err = register_pernet_device(&sit_net_ops);
1225         if (err < 0)
1226                 return err;
1227         err = xfrm4_tunnel_register(&sit_handler, AF_INET6);
1228         if (err < 0) {
1229                 unregister_pernet_device(&sit_net_ops);
1230                 printk(KERN_INFO "sit init: Can't add protocol\n");
1231         }
1232         return err;
1233 }
1234
1235 module_init(sit_init);
1236 module_exit(sit_cleanup);
1237 MODULE_LICENSE("GPL");
1238 MODULE_ALIAS("sit0");