Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net
[pandora-kernel.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 int ipip6_tunnel_init(struct net_device *dev);
67 static void ipip6_tunnel_setup(struct net_device *dev);
68 static void ipip6_dev_free(struct net_device *dev);
69
70 static int sit_net_id __read_mostly;
71 struct sit_net {
72         struct ip_tunnel __rcu *tunnels_r_l[HASH_SIZE];
73         struct ip_tunnel __rcu *tunnels_r[HASH_SIZE];
74         struct ip_tunnel __rcu *tunnels_l[HASH_SIZE];
75         struct ip_tunnel __rcu *tunnels_wc[1];
76         struct ip_tunnel __rcu **tunnels[4];
77
78         struct net_device *fb_tunnel_dev;
79 };
80
81 /*
82  * Locking : hash tables are protected by RCU and RTNL
83  */
84
85 #define for_each_ip_tunnel_rcu(start) \
86         for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
87
88 /* often modified stats are per cpu, other are shared (netdev->stats) */
89 struct pcpu_tstats {
90         unsigned long   rx_packets;
91         unsigned long   rx_bytes;
92         unsigned long   tx_packets;
93         unsigned long   tx_bytes;
94 };
95
96 static struct net_device_stats *ipip6_get_stats(struct net_device *dev)
97 {
98         struct pcpu_tstats sum = { 0 };
99         int i;
100
101         for_each_possible_cpu(i) {
102                 const struct pcpu_tstats *tstats = per_cpu_ptr(dev->tstats, i);
103
104                 sum.rx_packets += tstats->rx_packets;
105                 sum.rx_bytes   += tstats->rx_bytes;
106                 sum.tx_packets += tstats->tx_packets;
107                 sum.tx_bytes   += tstats->tx_bytes;
108         }
109         dev->stats.rx_packets = sum.rx_packets;
110         dev->stats.rx_bytes   = sum.rx_bytes;
111         dev->stats.tx_packets = sum.tx_packets;
112         dev->stats.tx_bytes   = sum.tx_bytes;
113         return &dev->stats;
114 }
115 /*
116  * Must be invoked with rcu_read_lock
117  */
118 static struct ip_tunnel * ipip6_tunnel_lookup(struct net *net,
119                 struct net_device *dev, __be32 remote, __be32 local)
120 {
121         unsigned int h0 = HASH(remote);
122         unsigned int h1 = HASH(local);
123         struct ip_tunnel *t;
124         struct sit_net *sitn = net_generic(net, sit_net_id);
125
126         for_each_ip_tunnel_rcu(sitn->tunnels_r_l[h0 ^ h1]) {
127                 if (local == t->parms.iph.saddr &&
128                     remote == t->parms.iph.daddr &&
129                     (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
130                     (t->dev->flags & IFF_UP))
131                         return t;
132         }
133         for_each_ip_tunnel_rcu(sitn->tunnels_r[h0]) {
134                 if (remote == t->parms.iph.daddr &&
135                     (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
136                     (t->dev->flags & IFF_UP))
137                         return t;
138         }
139         for_each_ip_tunnel_rcu(sitn->tunnels_l[h1]) {
140                 if (local == t->parms.iph.saddr &&
141                     (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
142                     (t->dev->flags & IFF_UP))
143                         return t;
144         }
145         t = rcu_dereference(sitn->tunnels_wc[0]);
146         if ((t != NULL) && (t->dev->flags & IFF_UP))
147                 return t;
148         return NULL;
149 }
150
151 static struct ip_tunnel __rcu **__ipip6_bucket(struct sit_net *sitn,
152                 struct ip_tunnel_parm *parms)
153 {
154         __be32 remote = parms->iph.daddr;
155         __be32 local = parms->iph.saddr;
156         unsigned int h = 0;
157         int prio = 0;
158
159         if (remote) {
160                 prio |= 2;
161                 h ^= HASH(remote);
162         }
163         if (local) {
164                 prio |= 1;
165                 h ^= HASH(local);
166         }
167         return &sitn->tunnels[prio][h];
168 }
169
170 static inline struct ip_tunnel __rcu **ipip6_bucket(struct sit_net *sitn,
171                 struct ip_tunnel *t)
172 {
173         return __ipip6_bucket(sitn, &t->parms);
174 }
175
176 static void ipip6_tunnel_unlink(struct sit_net *sitn, struct ip_tunnel *t)
177 {
178         struct ip_tunnel __rcu **tp;
179         struct ip_tunnel *iter;
180
181         for (tp = ipip6_bucket(sitn, t);
182              (iter = rtnl_dereference(*tp)) != NULL;
183              tp = &iter->next) {
184                 if (t == iter) {
185                         RCU_INIT_POINTER(*tp, t->next);
186                         break;
187                 }
188         }
189 }
190
191 static void ipip6_tunnel_link(struct sit_net *sitn, struct ip_tunnel *t)
192 {
193         struct ip_tunnel __rcu **tp = ipip6_bucket(sitn, t);
194
195         RCU_INIT_POINTER(t->next, rtnl_dereference(*tp));
196         RCU_INIT_POINTER(*tp, t);
197 }
198
199 static void ipip6_tunnel_clone_6rd(struct net_device *dev, struct sit_net *sitn)
200 {
201 #ifdef CONFIG_IPV6_SIT_6RD
202         struct ip_tunnel *t = netdev_priv(dev);
203
204         if (t->dev == sitn->fb_tunnel_dev) {
205                 ipv6_addr_set(&t->ip6rd.prefix, htonl(0x20020000), 0, 0, 0);
206                 t->ip6rd.relay_prefix = 0;
207                 t->ip6rd.prefixlen = 16;
208                 t->ip6rd.relay_prefixlen = 0;
209         } else {
210                 struct ip_tunnel *t0 = netdev_priv(sitn->fb_tunnel_dev);
211                 memcpy(&t->ip6rd, &t0->ip6rd, sizeof(t->ip6rd));
212         }
213 #endif
214 }
215
216 static struct ip_tunnel *ipip6_tunnel_locate(struct net *net,
217                 struct ip_tunnel_parm *parms, int create)
218 {
219         __be32 remote = parms->iph.daddr;
220         __be32 local = parms->iph.saddr;
221         struct ip_tunnel *t, *nt;
222         struct ip_tunnel __rcu **tp;
223         struct net_device *dev;
224         char name[IFNAMSIZ];
225         struct sit_net *sitn = net_generic(net, sit_net_id);
226
227         for (tp = __ipip6_bucket(sitn, parms);
228             (t = rtnl_dereference(*tp)) != NULL;
229              tp = &t->next) {
230                 if (local == t->parms.iph.saddr &&
231                     remote == t->parms.iph.daddr &&
232                     parms->link == t->parms.link) {
233                         if (create)
234                                 return NULL;
235                         else
236                                 return t;
237                 }
238         }
239         if (!create)
240                 goto failed;
241
242         if (parms->name[0])
243                 strlcpy(name, parms->name, IFNAMSIZ);
244         else
245                 strcpy(name, "sit%d");
246
247         dev = alloc_netdev(sizeof(*t), name, ipip6_tunnel_setup);
248         if (dev == NULL)
249                 return NULL;
250
251         dev_net_set(dev, net);
252
253         nt = netdev_priv(dev);
254
255         nt->parms = *parms;
256         if (ipip6_tunnel_init(dev) < 0)
257                 goto failed_free;
258         ipip6_tunnel_clone_6rd(dev, sitn);
259
260         if (parms->i_flags & SIT_ISATAP)
261                 dev->priv_flags |= IFF_ISATAP;
262
263         if (register_netdevice(dev) < 0)
264                 goto failed_free;
265
266         dev_hold(dev);
267
268         ipip6_tunnel_link(sitn, nt);
269         return nt;
270
271 failed_free:
272         ipip6_dev_free(dev);
273 failed:
274         return NULL;
275 }
276
277 #define for_each_prl_rcu(start)                 \
278         for (prl = rcu_dereference(start);      \
279              prl;                               \
280              prl = rcu_dereference(prl->next))
281
282 static struct ip_tunnel_prl_entry *
283 __ipip6_tunnel_locate_prl(struct ip_tunnel *t, __be32 addr)
284 {
285         struct ip_tunnel_prl_entry *prl;
286
287         for_each_prl_rcu(t->prl)
288                 if (prl->addr == addr)
289                         break;
290         return prl;
291
292 }
293
294 static int ipip6_tunnel_get_prl(struct ip_tunnel *t,
295                                 struct ip_tunnel_prl __user *a)
296 {
297         struct ip_tunnel_prl kprl, *kp;
298         struct ip_tunnel_prl_entry *prl;
299         unsigned int cmax, c = 0, ca, len;
300         int ret = 0;
301
302         if (copy_from_user(&kprl, a, sizeof(kprl)))
303                 return -EFAULT;
304         cmax = kprl.datalen / sizeof(kprl);
305         if (cmax > 1 && kprl.addr != htonl(INADDR_ANY))
306                 cmax = 1;
307
308         /* For simple GET or for root users,
309          * we try harder to allocate.
310          */
311         kp = (cmax <= 1 || capable(CAP_NET_ADMIN)) ?
312                 kcalloc(cmax, sizeof(*kp), GFP_KERNEL) :
313                 NULL;
314
315         rcu_read_lock();
316
317         ca = t->prl_count < cmax ? t->prl_count : cmax;
318
319         if (!kp) {
320                 /* We don't try hard to allocate much memory for
321                  * non-root users.
322                  * For root users, retry allocating enough memory for
323                  * the answer.
324                  */
325                 kp = kcalloc(ca, sizeof(*kp), GFP_ATOMIC);
326                 if (!kp) {
327                         ret = -ENOMEM;
328                         goto out;
329                 }
330         }
331
332         c = 0;
333         for_each_prl_rcu(t->prl) {
334                 if (c >= cmax)
335                         break;
336                 if (kprl.addr != htonl(INADDR_ANY) && prl->addr != kprl.addr)
337                         continue;
338                 kp[c].addr = prl->addr;
339                 kp[c].flags = prl->flags;
340                 c++;
341                 if (kprl.addr != htonl(INADDR_ANY))
342                         break;
343         }
344 out:
345         rcu_read_unlock();
346
347         len = sizeof(*kp) * c;
348         ret = 0;
349         if ((len && copy_to_user(a + 1, kp, len)) || put_user(len, &a->datalen))
350                 ret = -EFAULT;
351
352         kfree(kp);
353
354         return ret;
355 }
356
357 static int
358 ipip6_tunnel_add_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a, int chg)
359 {
360         struct ip_tunnel_prl_entry *p;
361         int err = 0;
362
363         if (a->addr == htonl(INADDR_ANY))
364                 return -EINVAL;
365
366         ASSERT_RTNL();
367
368         for (p = rtnl_dereference(t->prl); p; p = rtnl_dereference(p->next)) {
369                 if (p->addr == a->addr) {
370                         if (chg) {
371                                 p->flags = a->flags;
372                                 goto out;
373                         }
374                         err = -EEXIST;
375                         goto out;
376                 }
377         }
378
379         if (chg) {
380                 err = -ENXIO;
381                 goto out;
382         }
383
384         p = kzalloc(sizeof(struct ip_tunnel_prl_entry), GFP_KERNEL);
385         if (!p) {
386                 err = -ENOBUFS;
387                 goto out;
388         }
389
390         p->next = t->prl;
391         p->addr = a->addr;
392         p->flags = a->flags;
393         t->prl_count++;
394         RCU_INIT_POINTER(t->prl, p);
395 out:
396         return err;
397 }
398
399 static void prl_list_destroy_rcu(struct rcu_head *head)
400 {
401         struct ip_tunnel_prl_entry *p, *n;
402
403         p = container_of(head, struct ip_tunnel_prl_entry, rcu_head);
404         do {
405                 n = rcu_dereference_protected(p->next, 1);
406                 kfree(p);
407                 p = n;
408         } while (p);
409 }
410
411 static int
412 ipip6_tunnel_del_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a)
413 {
414         struct ip_tunnel_prl_entry *x;
415         struct ip_tunnel_prl_entry __rcu **p;
416         int err = 0;
417
418         ASSERT_RTNL();
419
420         if (a && a->addr != htonl(INADDR_ANY)) {
421                 for (p = &t->prl;
422                      (x = rtnl_dereference(*p)) != NULL;
423                      p = &x->next) {
424                         if (x->addr == a->addr) {
425                                 *p = x->next;
426                                 kfree_rcu(x, rcu_head);
427                                 t->prl_count--;
428                                 goto out;
429                         }
430                 }
431                 err = -ENXIO;
432         } else {
433                 x = rtnl_dereference(t->prl);
434                 if (x) {
435                         t->prl_count = 0;
436                         call_rcu(&x->rcu_head, prl_list_destroy_rcu);
437                         t->prl = NULL;
438                 }
439         }
440 out:
441         return err;
442 }
443
444 static int
445 isatap_chksrc(struct sk_buff *skb, const struct iphdr *iph, struct ip_tunnel *t)
446 {
447         struct ip_tunnel_prl_entry *p;
448         int ok = 1;
449
450         rcu_read_lock();
451         p = __ipip6_tunnel_locate_prl(t, iph->saddr);
452         if (p) {
453                 if (p->flags & PRL_DEFAULT)
454                         skb->ndisc_nodetype = NDISC_NODETYPE_DEFAULT;
455                 else
456                         skb->ndisc_nodetype = NDISC_NODETYPE_NODEFAULT;
457         } else {
458                 const struct in6_addr *addr6 = &ipv6_hdr(skb)->saddr;
459
460                 if (ipv6_addr_is_isatap(addr6) &&
461                     (addr6->s6_addr32[3] == iph->saddr) &&
462                     ipv6_chk_prefix(addr6, t->dev))
463                         skb->ndisc_nodetype = NDISC_NODETYPE_HOST;
464                 else
465                         ok = 0;
466         }
467         rcu_read_unlock();
468         return ok;
469 }
470
471 static void ipip6_tunnel_uninit(struct net_device *dev)
472 {
473         struct net *net = dev_net(dev);
474         struct sit_net *sitn = net_generic(net, sit_net_id);
475
476         if (dev == sitn->fb_tunnel_dev) {
477                 RCU_INIT_POINTER(sitn->tunnels_wc[0], NULL);
478         } else {
479                 ipip6_tunnel_unlink(sitn, netdev_priv(dev));
480                 ipip6_tunnel_del_prl(netdev_priv(dev), NULL);
481         }
482         dev_put(dev);
483 }
484
485
486 static int ipip6_err(struct sk_buff *skb, u32 info)
487 {
488
489 /* All the routers (except for Linux) return only
490    8 bytes of packet payload. It means, that precise relaying of
491    ICMP in the real Internet is absolutely infeasible.
492  */
493         const struct iphdr *iph = (const struct iphdr *)skb->data;
494         const int type = icmp_hdr(skb)->type;
495         const int code = icmp_hdr(skb)->code;
496         struct ip_tunnel *t;
497         int err;
498
499         switch (type) {
500         default:
501         case ICMP_PARAMETERPROB:
502                 return 0;
503
504         case ICMP_DEST_UNREACH:
505                 switch (code) {
506                 case ICMP_SR_FAILED:
507                 case ICMP_PORT_UNREACH:
508                         /* Impossible event. */
509                         return 0;
510                 case ICMP_FRAG_NEEDED:
511                         /* Soft state for pmtu is maintained by IP core. */
512                         return 0;
513                 default:
514                         /* All others are translated to HOST_UNREACH.
515                            rfc2003 contains "deep thoughts" about NET_UNREACH,
516                            I believe they are just ether pollution. --ANK
517                          */
518                         break;
519                 }
520                 break;
521         case ICMP_TIME_EXCEEDED:
522                 if (code != ICMP_EXC_TTL)
523                         return 0;
524                 break;
525         }
526
527         err = -ENOENT;
528
529         rcu_read_lock();
530         t = ipip6_tunnel_lookup(dev_net(skb->dev),
531                                 skb->dev,
532                                 iph->daddr,
533                                 iph->saddr);
534         if (t == NULL || t->parms.iph.daddr == 0)
535                 goto out;
536
537         err = 0;
538         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
539                 goto out;
540
541         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
542                 t->err_count++;
543         else
544                 t->err_count = 1;
545         t->err_time = jiffies;
546 out:
547         rcu_read_unlock();
548         return err;
549 }
550
551 static inline void ipip6_ecn_decapsulate(const struct iphdr *iph, struct sk_buff *skb)
552 {
553         if (INET_ECN_is_ce(iph->tos))
554                 IP6_ECN_set_ce(ipv6_hdr(skb));
555 }
556
557 static int ipip6_rcv(struct sk_buff *skb)
558 {
559         const struct iphdr *iph;
560         struct ip_tunnel *tunnel;
561
562         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
563                 goto out;
564
565         iph = ip_hdr(skb);
566
567         rcu_read_lock();
568         tunnel = ipip6_tunnel_lookup(dev_net(skb->dev), skb->dev,
569                                      iph->saddr, iph->daddr);
570         if (tunnel != NULL) {
571                 struct pcpu_tstats *tstats;
572
573                 secpath_reset(skb);
574                 skb->mac_header = skb->network_header;
575                 skb_reset_network_header(skb);
576                 IPCB(skb)->flags = 0;
577                 skb->protocol = htons(ETH_P_IPV6);
578                 skb->pkt_type = PACKET_HOST;
579
580                 if ((tunnel->dev->priv_flags & IFF_ISATAP) &&
581                     !isatap_chksrc(skb, iph, tunnel)) {
582                         tunnel->dev->stats.rx_errors++;
583                         rcu_read_unlock();
584                         kfree_skb(skb);
585                         return 0;
586                 }
587
588                 tstats = this_cpu_ptr(tunnel->dev->tstats);
589                 tstats->rx_packets++;
590                 tstats->rx_bytes += skb->len;
591
592                 __skb_tunnel_rx(skb, tunnel->dev);
593
594                 ipip6_ecn_decapsulate(iph, skb);
595
596                 netif_rx(skb);
597
598                 rcu_read_unlock();
599                 return 0;
600         }
601
602         /* no tunnel matched,  let upstream know, ipsec may handle it */
603         rcu_read_unlock();
604         return 1;
605 out:
606         kfree_skb(skb);
607         return 0;
608 }
609
610 /*
611  * Returns the embedded IPv4 address if the IPv6 address
612  * comes from 6rd / 6to4 (RFC 3056) addr space.
613  */
614 static inline
615 __be32 try_6rd(const struct in6_addr *v6dst, struct ip_tunnel *tunnel)
616 {
617         __be32 dst = 0;
618
619 #ifdef CONFIG_IPV6_SIT_6RD
620         if (ipv6_prefix_equal(v6dst, &tunnel->ip6rd.prefix,
621                               tunnel->ip6rd.prefixlen)) {
622                 unsigned int pbw0, pbi0;
623                 int pbi1;
624                 u32 d;
625
626                 pbw0 = tunnel->ip6rd.prefixlen >> 5;
627                 pbi0 = tunnel->ip6rd.prefixlen & 0x1f;
628
629                 d = (ntohl(v6dst->s6_addr32[pbw0]) << pbi0) >>
630                     tunnel->ip6rd.relay_prefixlen;
631
632                 pbi1 = pbi0 - tunnel->ip6rd.relay_prefixlen;
633                 if (pbi1 > 0)
634                         d |= ntohl(v6dst->s6_addr32[pbw0 + 1]) >>
635                              (32 - pbi1);
636
637                 dst = tunnel->ip6rd.relay_prefix | htonl(d);
638         }
639 #else
640         if (v6dst->s6_addr16[0] == htons(0x2002)) {
641                 /* 6to4 v6 addr has 16 bits prefix, 32 v4addr, 16 SLA, ... */
642                 memcpy(&dst, &v6dst->s6_addr16[1], 4);
643         }
644 #endif
645         return dst;
646 }
647
648 /*
649  *      This function assumes it is being called from dev_queue_xmit()
650  *      and that skb is filled properly by that function.
651  */
652
653 static netdev_tx_t ipip6_tunnel_xmit(struct sk_buff *skb,
654                                      struct net_device *dev)
655 {
656         struct ip_tunnel *tunnel = netdev_priv(dev);
657         struct pcpu_tstats *tstats;
658         const struct iphdr  *tiph = &tunnel->parms.iph;
659         const struct ipv6hdr *iph6 = ipv6_hdr(skb);
660         u8     tos = tunnel->parms.iph.tos;
661         __be16 df = tiph->frag_off;
662         struct rtable *rt;                      /* Route to the other host */
663         struct net_device *tdev;                /* Device to other host */
664         struct iphdr  *iph;                     /* Our new IP header */
665         unsigned int max_headroom;              /* The extra header space needed */
666         __be32 dst = tiph->daddr;
667         struct flowi4 fl4;
668         int    mtu;
669         const struct in6_addr *addr6;
670         int addr_type;
671
672         if (skb->protocol != htons(ETH_P_IPV6))
673                 goto tx_error;
674
675         if (tos == 1)
676                 tos = ipv6_get_dsfield(iph6);
677
678         /* ISATAP (RFC4214) - must come before 6to4 */
679         if (dev->priv_flags & IFF_ISATAP) {
680                 struct neighbour *neigh = NULL;
681
682                 if (skb_dst(skb))
683                         neigh = dst_get_neighbour(skb_dst(skb));
684
685                 if (neigh == NULL) {
686                         if (net_ratelimit())
687                                 printk(KERN_DEBUG "sit: nexthop == NULL\n");
688                         goto tx_error;
689                 }
690
691                 addr6 = (const struct in6_addr*)&neigh->primary_key;
692                 addr_type = ipv6_addr_type(addr6);
693
694                 if ((addr_type & IPV6_ADDR_UNICAST) &&
695                      ipv6_addr_is_isatap(addr6))
696                         dst = addr6->s6_addr32[3];
697                 else
698                         goto tx_error;
699         }
700
701         if (!dst)
702                 dst = try_6rd(&iph6->daddr, tunnel);
703
704         if (!dst) {
705                 struct neighbour *neigh = NULL;
706
707                 if (skb_dst(skb))
708                         neigh = dst_get_neighbour(skb_dst(skb));
709
710                 if (neigh == NULL) {
711                         if (net_ratelimit())
712                                 printk(KERN_DEBUG "sit: nexthop == NULL\n");
713                         goto tx_error;
714                 }
715
716                 addr6 = (const struct in6_addr*)&neigh->primary_key;
717                 addr_type = ipv6_addr_type(addr6);
718
719                 if (addr_type == IPV6_ADDR_ANY) {
720                         addr6 = &ipv6_hdr(skb)->daddr;
721                         addr_type = ipv6_addr_type(addr6);
722                 }
723
724                 if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
725                         goto tx_error_icmp;
726
727                 dst = addr6->s6_addr32[3];
728         }
729
730         rt = ip_route_output_ports(dev_net(dev), &fl4, NULL,
731                                    dst, tiph->saddr,
732                                    0, 0,
733                                    IPPROTO_IPV6, RT_TOS(tos),
734                                    tunnel->parms.link);
735         if (IS_ERR(rt)) {
736                 dev->stats.tx_carrier_errors++;
737                 goto tx_error_icmp;
738         }
739         if (rt->rt_type != RTN_UNICAST) {
740                 ip_rt_put(rt);
741                 dev->stats.tx_carrier_errors++;
742                 goto tx_error_icmp;
743         }
744         tdev = rt->dst.dev;
745
746         if (tdev == dev) {
747                 ip_rt_put(rt);
748                 dev->stats.collisions++;
749                 goto tx_error;
750         }
751
752         if (df) {
753                 mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
754
755                 if (mtu < 68) {
756                         dev->stats.collisions++;
757                         ip_rt_put(rt);
758                         goto tx_error;
759                 }
760
761                 if (mtu < IPV6_MIN_MTU) {
762                         mtu = IPV6_MIN_MTU;
763                         df = 0;
764                 }
765
766                 if (tunnel->parms.iph.daddr && skb_dst(skb))
767                         skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
768
769                 if (skb->len > mtu) {
770                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
771                         ip_rt_put(rt);
772                         goto tx_error;
773                 }
774         }
775
776         if (tunnel->err_count > 0) {
777                 if (time_before(jiffies,
778                                 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
779                         tunnel->err_count--;
780                         dst_link_failure(skb);
781                 } else
782                         tunnel->err_count = 0;
783         }
784
785         /*
786          * Okay, now see if we can stuff it in the buffer as-is.
787          */
788         max_headroom = LL_RESERVED_SPACE(tdev)+sizeof(struct iphdr);
789
790         if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
791             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
792                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
793                 if (!new_skb) {
794                         ip_rt_put(rt);
795                         dev->stats.tx_dropped++;
796                         dev_kfree_skb(skb);
797                         return NETDEV_TX_OK;
798                 }
799                 if (skb->sk)
800                         skb_set_owner_w(new_skb, skb->sk);
801                 dev_kfree_skb(skb);
802                 skb = new_skb;
803                 iph6 = ipv6_hdr(skb);
804         }
805
806         skb->transport_header = skb->network_header;
807         skb_push(skb, sizeof(struct iphdr));
808         skb_reset_network_header(skb);
809         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
810         IPCB(skb)->flags = 0;
811         skb_dst_drop(skb);
812         skb_dst_set(skb, &rt->dst);
813
814         /*
815          *      Push down and install the IPIP header.
816          */
817
818         iph                     =       ip_hdr(skb);
819         iph->version            =       4;
820         iph->ihl                =       sizeof(struct iphdr)>>2;
821         iph->frag_off           =       df;
822         iph->protocol           =       IPPROTO_IPV6;
823         iph->tos                =       INET_ECN_encapsulate(tos, ipv6_get_dsfield(iph6));
824         iph->daddr              =       fl4.daddr;
825         iph->saddr              =       fl4.saddr;
826
827         if ((iph->ttl = tiph->ttl) == 0)
828                 iph->ttl        =       iph6->hop_limit;
829
830         nf_reset(skb);
831         tstats = this_cpu_ptr(dev->tstats);
832         __IPTUNNEL_XMIT(tstats, &dev->stats);
833         return NETDEV_TX_OK;
834
835 tx_error_icmp:
836         dst_link_failure(skb);
837 tx_error:
838         dev->stats.tx_errors++;
839         dev_kfree_skb(skb);
840         return NETDEV_TX_OK;
841 }
842
843 static void ipip6_tunnel_bind_dev(struct net_device *dev)
844 {
845         struct net_device *tdev = NULL;
846         struct ip_tunnel *tunnel;
847         const struct iphdr *iph;
848         struct flowi4 fl4;
849
850         tunnel = netdev_priv(dev);
851         iph = &tunnel->parms.iph;
852
853         if (iph->daddr) {
854                 struct rtable *rt = ip_route_output_ports(dev_net(dev), &fl4, NULL,
855                                                           iph->daddr, iph->saddr,
856                                                           0, 0,
857                                                           IPPROTO_IPV6,
858                                                           RT_TOS(iph->tos),
859                                                           tunnel->parms.link);
860
861                 if (!IS_ERR(rt)) {
862                         tdev = rt->dst.dev;
863                         ip_rt_put(rt);
864                 }
865                 dev->flags |= IFF_POINTOPOINT;
866         }
867
868         if (!tdev && tunnel->parms.link)
869                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
870
871         if (tdev) {
872                 dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
873                 dev->mtu = tdev->mtu - sizeof(struct iphdr);
874                 if (dev->mtu < IPV6_MIN_MTU)
875                         dev->mtu = IPV6_MIN_MTU;
876         }
877         dev->iflink = tunnel->parms.link;
878 }
879
880 static int
881 ipip6_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
882 {
883         int err = 0;
884         struct ip_tunnel_parm p;
885         struct ip_tunnel_prl prl;
886         struct ip_tunnel *t;
887         struct net *net = dev_net(dev);
888         struct sit_net *sitn = net_generic(net, sit_net_id);
889 #ifdef CONFIG_IPV6_SIT_6RD
890         struct ip_tunnel_6rd ip6rd;
891 #endif
892
893         switch (cmd) {
894         case SIOCGETTUNNEL:
895 #ifdef CONFIG_IPV6_SIT_6RD
896         case SIOCGET6RD:
897 #endif
898                 t = NULL;
899                 if (dev == sitn->fb_tunnel_dev) {
900                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
901                                 err = -EFAULT;
902                                 break;
903                         }
904                         t = ipip6_tunnel_locate(net, &p, 0);
905                 }
906                 if (t == NULL)
907                         t = netdev_priv(dev);
908
909                 err = -EFAULT;
910                 if (cmd == SIOCGETTUNNEL) {
911                         memcpy(&p, &t->parms, sizeof(p));
912                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &p,
913                                          sizeof(p)))
914                                 goto done;
915 #ifdef CONFIG_IPV6_SIT_6RD
916                 } else {
917                         ipv6_addr_copy(&ip6rd.prefix, &t->ip6rd.prefix);
918                         ip6rd.relay_prefix = t->ip6rd.relay_prefix;
919                         ip6rd.prefixlen = t->ip6rd.prefixlen;
920                         ip6rd.relay_prefixlen = t->ip6rd.relay_prefixlen;
921                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &ip6rd,
922                                          sizeof(ip6rd)))
923                                 goto done;
924 #endif
925                 }
926                 err = 0;
927                 break;
928
929         case SIOCADDTUNNEL:
930         case SIOCCHGTUNNEL:
931                 err = -EPERM;
932                 if (!capable(CAP_NET_ADMIN))
933                         goto done;
934
935                 err = -EFAULT;
936                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
937                         goto done;
938
939                 err = -EINVAL;
940                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPV6 ||
941                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
942                         goto done;
943                 if (p.iph.ttl)
944                         p.iph.frag_off |= htons(IP_DF);
945
946                 t = ipip6_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
947
948                 if (dev != sitn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
949                         if (t != NULL) {
950                                 if (t->dev != dev) {
951                                         err = -EEXIST;
952                                         break;
953                                 }
954                         } else {
955                                 if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
956                                     (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
957                                         err = -EINVAL;
958                                         break;
959                                 }
960                                 t = netdev_priv(dev);
961                                 ipip6_tunnel_unlink(sitn, t);
962                                 synchronize_net();
963                                 t->parms.iph.saddr = p.iph.saddr;
964                                 t->parms.iph.daddr = p.iph.daddr;
965                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
966                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
967                                 ipip6_tunnel_link(sitn, t);
968                                 netdev_state_change(dev);
969                         }
970                 }
971
972                 if (t) {
973                         err = 0;
974                         if (cmd == SIOCCHGTUNNEL) {
975                                 t->parms.iph.ttl = p.iph.ttl;
976                                 t->parms.iph.tos = p.iph.tos;
977                                 if (t->parms.link != p.link) {
978                                         t->parms.link = p.link;
979                                         ipip6_tunnel_bind_dev(dev);
980                                         netdev_state_change(dev);
981                                 }
982                         }
983                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
984                                 err = -EFAULT;
985                 } else
986                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
987                 break;
988
989         case SIOCDELTUNNEL:
990                 err = -EPERM;
991                 if (!capable(CAP_NET_ADMIN))
992                         goto done;
993
994                 if (dev == sitn->fb_tunnel_dev) {
995                         err = -EFAULT;
996                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
997                                 goto done;
998                         err = -ENOENT;
999                         if ((t = ipip6_tunnel_locate(net, &p, 0)) == NULL)
1000                                 goto done;
1001                         err = -EPERM;
1002                         if (t == netdev_priv(sitn->fb_tunnel_dev))
1003                                 goto done;
1004                         dev = t->dev;
1005                 }
1006                 unregister_netdevice(dev);
1007                 err = 0;
1008                 break;
1009
1010         case SIOCGETPRL:
1011                 err = -EINVAL;
1012                 if (dev == sitn->fb_tunnel_dev)
1013                         goto done;
1014                 err = -ENOENT;
1015                 if (!(t = netdev_priv(dev)))
1016                         goto done;
1017                 err = ipip6_tunnel_get_prl(t, ifr->ifr_ifru.ifru_data);
1018                 break;
1019
1020         case SIOCADDPRL:
1021         case SIOCDELPRL:
1022         case SIOCCHGPRL:
1023                 err = -EPERM;
1024                 if (!capable(CAP_NET_ADMIN))
1025                         goto done;
1026                 err = -EINVAL;
1027                 if (dev == sitn->fb_tunnel_dev)
1028                         goto done;
1029                 err = -EFAULT;
1030                 if (copy_from_user(&prl, ifr->ifr_ifru.ifru_data, sizeof(prl)))
1031                         goto done;
1032                 err = -ENOENT;
1033                 if (!(t = netdev_priv(dev)))
1034                         goto done;
1035
1036                 switch (cmd) {
1037                 case SIOCDELPRL:
1038                         err = ipip6_tunnel_del_prl(t, &prl);
1039                         break;
1040                 case SIOCADDPRL:
1041                 case SIOCCHGPRL:
1042                         err = ipip6_tunnel_add_prl(t, &prl, cmd == SIOCCHGPRL);
1043                         break;
1044                 }
1045                 netdev_state_change(dev);
1046                 break;
1047
1048 #ifdef CONFIG_IPV6_SIT_6RD
1049         case SIOCADD6RD:
1050         case SIOCCHG6RD:
1051         case SIOCDEL6RD:
1052                 err = -EPERM;
1053                 if (!capable(CAP_NET_ADMIN))
1054                         goto done;
1055
1056                 err = -EFAULT;
1057                 if (copy_from_user(&ip6rd, ifr->ifr_ifru.ifru_data,
1058                                    sizeof(ip6rd)))
1059                         goto done;
1060
1061                 t = netdev_priv(dev);
1062
1063                 if (cmd != SIOCDEL6RD) {
1064                         struct in6_addr prefix;
1065                         __be32 relay_prefix;
1066
1067                         err = -EINVAL;
1068                         if (ip6rd.relay_prefixlen > 32 ||
1069                             ip6rd.prefixlen + (32 - ip6rd.relay_prefixlen) > 64)
1070                                 goto done;
1071
1072                         ipv6_addr_prefix(&prefix, &ip6rd.prefix,
1073                                          ip6rd.prefixlen);
1074                         if (!ipv6_addr_equal(&prefix, &ip6rd.prefix))
1075                                 goto done;
1076                         if (ip6rd.relay_prefixlen)
1077                                 relay_prefix = ip6rd.relay_prefix &
1078                                                htonl(0xffffffffUL <<
1079                                                      (32 - ip6rd.relay_prefixlen));
1080                         else
1081                                 relay_prefix = 0;
1082                         if (relay_prefix != ip6rd.relay_prefix)
1083                                 goto done;
1084
1085                         ipv6_addr_copy(&t->ip6rd.prefix, &prefix);
1086                         t->ip6rd.relay_prefix = relay_prefix;
1087                         t->ip6rd.prefixlen = ip6rd.prefixlen;
1088                         t->ip6rd.relay_prefixlen = ip6rd.relay_prefixlen;
1089                 } else
1090                         ipip6_tunnel_clone_6rd(dev, sitn);
1091
1092                 err = 0;
1093                 break;
1094 #endif
1095
1096         default:
1097                 err = -EINVAL;
1098         }
1099
1100 done:
1101         return err;
1102 }
1103
1104 static int ipip6_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1105 {
1106         if (new_mtu < IPV6_MIN_MTU || new_mtu > 0xFFF8 - sizeof(struct iphdr))
1107                 return -EINVAL;
1108         dev->mtu = new_mtu;
1109         return 0;
1110 }
1111
1112 static const struct net_device_ops ipip6_netdev_ops = {
1113         .ndo_uninit     = ipip6_tunnel_uninit,
1114         .ndo_start_xmit = ipip6_tunnel_xmit,
1115         .ndo_do_ioctl   = ipip6_tunnel_ioctl,
1116         .ndo_change_mtu = ipip6_tunnel_change_mtu,
1117         .ndo_get_stats  = ipip6_get_stats,
1118 };
1119
1120 static void ipip6_dev_free(struct net_device *dev)
1121 {
1122         free_percpu(dev->tstats);
1123         free_netdev(dev);
1124 }
1125
1126 static void ipip6_tunnel_setup(struct net_device *dev)
1127 {
1128         dev->netdev_ops         = &ipip6_netdev_ops;
1129         dev->destructor         = ipip6_dev_free;
1130
1131         dev->type               = ARPHRD_SIT;
1132         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr);
1133         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr);
1134         dev->flags              = IFF_NOARP;
1135         dev->priv_flags        &= ~IFF_XMIT_DST_RELEASE;
1136         dev->iflink             = 0;
1137         dev->addr_len           = 4;
1138         dev->features           |= NETIF_F_NETNS_LOCAL;
1139         dev->features           |= NETIF_F_LLTX;
1140 }
1141
1142 static int ipip6_tunnel_init(struct net_device *dev)
1143 {
1144         struct ip_tunnel *tunnel = netdev_priv(dev);
1145
1146         tunnel->dev = dev;
1147         strcpy(tunnel->parms.name, dev->name);
1148
1149         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1150         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1151
1152         ipip6_tunnel_bind_dev(dev);
1153         dev->tstats = alloc_percpu(struct pcpu_tstats);
1154         if (!dev->tstats)
1155                 return -ENOMEM;
1156
1157         return 0;
1158 }
1159
1160 static int __net_init ipip6_fb_tunnel_init(struct net_device *dev)
1161 {
1162         struct ip_tunnel *tunnel = netdev_priv(dev);
1163         struct iphdr *iph = &tunnel->parms.iph;
1164         struct net *net = dev_net(dev);
1165         struct sit_net *sitn = net_generic(net, sit_net_id);
1166
1167         tunnel->dev = dev;
1168         strcpy(tunnel->parms.name, dev->name);
1169
1170         iph->version            = 4;
1171         iph->protocol           = IPPROTO_IPV6;
1172         iph->ihl                = 5;
1173         iph->ttl                = 64;
1174
1175         dev->tstats = alloc_percpu(struct pcpu_tstats);
1176         if (!dev->tstats)
1177                 return -ENOMEM;
1178         dev_hold(dev);
1179         RCU_INIT_POINTER(sitn->tunnels_wc[0], tunnel);
1180         return 0;
1181 }
1182
1183 static struct xfrm_tunnel sit_handler __read_mostly = {
1184         .handler        =       ipip6_rcv,
1185         .err_handler    =       ipip6_err,
1186         .priority       =       1,
1187 };
1188
1189 static void __net_exit sit_destroy_tunnels(struct sit_net *sitn, struct list_head *head)
1190 {
1191         int prio;
1192
1193         for (prio = 1; prio < 4; prio++) {
1194                 int h;
1195                 for (h = 0; h < HASH_SIZE; h++) {
1196                         struct ip_tunnel *t;
1197
1198                         t = rtnl_dereference(sitn->tunnels[prio][h]);
1199                         while (t != NULL) {
1200                                 unregister_netdevice_queue(t->dev, head);
1201                                 t = rtnl_dereference(t->next);
1202                         }
1203                 }
1204         }
1205 }
1206
1207 static int __net_init sit_init_net(struct net *net)
1208 {
1209         struct sit_net *sitn = net_generic(net, sit_net_id);
1210         int err;
1211
1212         sitn->tunnels[0] = sitn->tunnels_wc;
1213         sitn->tunnels[1] = sitn->tunnels_l;
1214         sitn->tunnels[2] = sitn->tunnels_r;
1215         sitn->tunnels[3] = sitn->tunnels_r_l;
1216
1217         sitn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "sit0",
1218                                            ipip6_tunnel_setup);
1219         if (!sitn->fb_tunnel_dev) {
1220                 err = -ENOMEM;
1221                 goto err_alloc_dev;
1222         }
1223         dev_net_set(sitn->fb_tunnel_dev, net);
1224
1225         err = ipip6_fb_tunnel_init(sitn->fb_tunnel_dev);
1226         if (err)
1227                 goto err_dev_free;
1228
1229         ipip6_tunnel_clone_6rd(sitn->fb_tunnel_dev, sitn);
1230
1231         if ((err = register_netdev(sitn->fb_tunnel_dev)))
1232                 goto err_reg_dev;
1233
1234         return 0;
1235
1236 err_reg_dev:
1237         dev_put(sitn->fb_tunnel_dev);
1238 err_dev_free:
1239         ipip6_dev_free(sitn->fb_tunnel_dev);
1240 err_alloc_dev:
1241         return err;
1242 }
1243
1244 static void __net_exit sit_exit_net(struct net *net)
1245 {
1246         struct sit_net *sitn = net_generic(net, sit_net_id);
1247         LIST_HEAD(list);
1248
1249         rtnl_lock();
1250         sit_destroy_tunnels(sitn, &list);
1251         unregister_netdevice_queue(sitn->fb_tunnel_dev, &list);
1252         unregister_netdevice_many(&list);
1253         rtnl_unlock();
1254 }
1255
1256 static struct pernet_operations sit_net_ops = {
1257         .init = sit_init_net,
1258         .exit = sit_exit_net,
1259         .id   = &sit_net_id,
1260         .size = sizeof(struct sit_net),
1261 };
1262
1263 static void __exit sit_cleanup(void)
1264 {
1265         xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
1266
1267         unregister_pernet_device(&sit_net_ops);
1268         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1269 }
1270
1271 static int __init sit_init(void)
1272 {
1273         int err;
1274
1275         printk(KERN_INFO "IPv6 over IPv4 tunneling driver\n");
1276
1277         err = register_pernet_device(&sit_net_ops);
1278         if (err < 0)
1279                 return err;
1280         err = xfrm4_tunnel_register(&sit_handler, AF_INET6);
1281         if (err < 0) {
1282                 unregister_pernet_device(&sit_net_ops);
1283                 printk(KERN_INFO "sit init: Can't add protocol\n");
1284         }
1285         return err;
1286 }
1287
1288 module_init(sit_init);
1289 module_exit(sit_cleanup);
1290 MODULE_LICENSE("GPL");
1291 MODULE_ALIAS_NETDEV("sit0");