net,rcu: convert call_rcu(prl_entry_destroy_rcu) to kfree
[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_assign_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_assign_pointer(t->next, rtnl_dereference(*tp));
196         rcu_assign_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         if (strchr(name, '%')) {
254                 if (dev_alloc_name(dev, name) < 0)
255                         goto failed_free;
256         }
257
258         nt = netdev_priv(dev);
259
260         nt->parms = *parms;
261         if (ipip6_tunnel_init(dev) < 0)
262                 goto failed_free;
263         ipip6_tunnel_clone_6rd(dev, sitn);
264
265         if (parms->i_flags & SIT_ISATAP)
266                 dev->priv_flags |= IFF_ISATAP;
267
268         if (register_netdevice(dev) < 0)
269                 goto failed_free;
270
271         dev_hold(dev);
272
273         ipip6_tunnel_link(sitn, nt);
274         return nt;
275
276 failed_free:
277         ipip6_dev_free(dev);
278 failed:
279         return NULL;
280 }
281
282 #define for_each_prl_rcu(start)                 \
283         for (prl = rcu_dereference(start);      \
284              prl;                               \
285              prl = rcu_dereference(prl->next))
286
287 static struct ip_tunnel_prl_entry *
288 __ipip6_tunnel_locate_prl(struct ip_tunnel *t, __be32 addr)
289 {
290         struct ip_tunnel_prl_entry *prl;
291
292         for_each_prl_rcu(t->prl)
293                 if (prl->addr == addr)
294                         break;
295         return prl;
296
297 }
298
299 static int ipip6_tunnel_get_prl(struct ip_tunnel *t,
300                                 struct ip_tunnel_prl __user *a)
301 {
302         struct ip_tunnel_prl kprl, *kp;
303         struct ip_tunnel_prl_entry *prl;
304         unsigned int cmax, c = 0, ca, len;
305         int ret = 0;
306
307         if (copy_from_user(&kprl, a, sizeof(kprl)))
308                 return -EFAULT;
309         cmax = kprl.datalen / sizeof(kprl);
310         if (cmax > 1 && kprl.addr != htonl(INADDR_ANY))
311                 cmax = 1;
312
313         /* For simple GET or for root users,
314          * we try harder to allocate.
315          */
316         kp = (cmax <= 1 || capable(CAP_NET_ADMIN)) ?
317                 kcalloc(cmax, sizeof(*kp), GFP_KERNEL) :
318                 NULL;
319
320         rcu_read_lock();
321
322         ca = t->prl_count < cmax ? t->prl_count : cmax;
323
324         if (!kp) {
325                 /* We don't try hard to allocate much memory for
326                  * non-root users.
327                  * For root users, retry allocating enough memory for
328                  * the answer.
329                  */
330                 kp = kcalloc(ca, sizeof(*kp), GFP_ATOMIC);
331                 if (!kp) {
332                         ret = -ENOMEM;
333                         goto out;
334                 }
335         }
336
337         c = 0;
338         for_each_prl_rcu(t->prl) {
339                 if (c >= cmax)
340                         break;
341                 if (kprl.addr != htonl(INADDR_ANY) && prl->addr != kprl.addr)
342                         continue;
343                 kp[c].addr = prl->addr;
344                 kp[c].flags = prl->flags;
345                 c++;
346                 if (kprl.addr != htonl(INADDR_ANY))
347                         break;
348         }
349 out:
350         rcu_read_unlock();
351
352         len = sizeof(*kp) * c;
353         ret = 0;
354         if ((len && copy_to_user(a + 1, kp, len)) || put_user(len, &a->datalen))
355                 ret = -EFAULT;
356
357         kfree(kp);
358
359         return ret;
360 }
361
362 static int
363 ipip6_tunnel_add_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a, int chg)
364 {
365         struct ip_tunnel_prl_entry *p;
366         int err = 0;
367
368         if (a->addr == htonl(INADDR_ANY))
369                 return -EINVAL;
370
371         ASSERT_RTNL();
372
373         for (p = rtnl_dereference(t->prl); p; p = rtnl_dereference(p->next)) {
374                 if (p->addr == a->addr) {
375                         if (chg) {
376                                 p->flags = a->flags;
377                                 goto out;
378                         }
379                         err = -EEXIST;
380                         goto out;
381                 }
382         }
383
384         if (chg) {
385                 err = -ENXIO;
386                 goto out;
387         }
388
389         p = kzalloc(sizeof(struct ip_tunnel_prl_entry), GFP_KERNEL);
390         if (!p) {
391                 err = -ENOBUFS;
392                 goto out;
393         }
394
395         p->next = t->prl;
396         p->addr = a->addr;
397         p->flags = a->flags;
398         t->prl_count++;
399         rcu_assign_pointer(t->prl, p);
400 out:
401         return err;
402 }
403
404 static void prl_list_destroy_rcu(struct rcu_head *head)
405 {
406         struct ip_tunnel_prl_entry *p, *n;
407
408         p = container_of(head, struct ip_tunnel_prl_entry, rcu_head);
409         do {
410                 n = rcu_dereference_protected(p->next, 1);
411                 kfree(p);
412                 p = n;
413         } while (p);
414 }
415
416 static int
417 ipip6_tunnel_del_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a)
418 {
419         struct ip_tunnel_prl_entry *x;
420         struct ip_tunnel_prl_entry __rcu **p;
421         int err = 0;
422
423         ASSERT_RTNL();
424
425         if (a && a->addr != htonl(INADDR_ANY)) {
426                 for (p = &t->prl;
427                      (x = rtnl_dereference(*p)) != NULL;
428                      p = &x->next) {
429                         if (x->addr == a->addr) {
430                                 *p = x->next;
431                                 kfree_rcu(x, rcu_head);
432                                 t->prl_count--;
433                                 goto out;
434                         }
435                 }
436                 err = -ENXIO;
437         } else {
438                 x = rtnl_dereference(t->prl);
439                 if (x) {
440                         t->prl_count = 0;
441                         call_rcu(&x->rcu_head, prl_list_destroy_rcu);
442                         t->prl = NULL;
443                 }
444         }
445 out:
446         return err;
447 }
448
449 static int
450 isatap_chksrc(struct sk_buff *skb, struct iphdr *iph, struct ip_tunnel *t)
451 {
452         struct ip_tunnel_prl_entry *p;
453         int ok = 1;
454
455         rcu_read_lock();
456         p = __ipip6_tunnel_locate_prl(t, iph->saddr);
457         if (p) {
458                 if (p->flags & PRL_DEFAULT)
459                         skb->ndisc_nodetype = NDISC_NODETYPE_DEFAULT;
460                 else
461                         skb->ndisc_nodetype = NDISC_NODETYPE_NODEFAULT;
462         } else {
463                 struct in6_addr *addr6 = &ipv6_hdr(skb)->saddr;
464                 if (ipv6_addr_is_isatap(addr6) &&
465                     (addr6->s6_addr32[3] == iph->saddr) &&
466                     ipv6_chk_prefix(addr6, t->dev))
467                         skb->ndisc_nodetype = NDISC_NODETYPE_HOST;
468                 else
469                         ok = 0;
470         }
471         rcu_read_unlock();
472         return ok;
473 }
474
475 static void ipip6_tunnel_uninit(struct net_device *dev)
476 {
477         struct net *net = dev_net(dev);
478         struct sit_net *sitn = net_generic(net, sit_net_id);
479
480         if (dev == sitn->fb_tunnel_dev) {
481                 rcu_assign_pointer(sitn->tunnels_wc[0], NULL);
482         } else {
483                 ipip6_tunnel_unlink(sitn, netdev_priv(dev));
484                 ipip6_tunnel_del_prl(netdev_priv(dev), NULL);
485         }
486         dev_put(dev);
487 }
488
489
490 static int ipip6_err(struct sk_buff *skb, u32 info)
491 {
492
493 /* All the routers (except for Linux) return only
494    8 bytes of packet payload. It means, that precise relaying of
495    ICMP in the real Internet is absolutely infeasible.
496  */
497         struct iphdr *iph = (struct iphdr*)skb->data;
498         const int type = icmp_hdr(skb)->type;
499         const int code = icmp_hdr(skb)->code;
500         struct ip_tunnel *t;
501         int err;
502
503         switch (type) {
504         default:
505         case ICMP_PARAMETERPROB:
506                 return 0;
507
508         case ICMP_DEST_UNREACH:
509                 switch (code) {
510                 case ICMP_SR_FAILED:
511                 case ICMP_PORT_UNREACH:
512                         /* Impossible event. */
513                         return 0;
514                 case ICMP_FRAG_NEEDED:
515                         /* Soft state for pmtu is maintained by IP core. */
516                         return 0;
517                 default:
518                         /* All others are translated to HOST_UNREACH.
519                            rfc2003 contains "deep thoughts" about NET_UNREACH,
520                            I believe they are just ether pollution. --ANK
521                          */
522                         break;
523                 }
524                 break;
525         case ICMP_TIME_EXCEEDED:
526                 if (code != ICMP_EXC_TTL)
527                         return 0;
528                 break;
529         }
530
531         err = -ENOENT;
532
533         rcu_read_lock();
534         t = ipip6_tunnel_lookup(dev_net(skb->dev),
535                                 skb->dev,
536                                 iph->daddr,
537                                 iph->saddr);
538         if (t == NULL || t->parms.iph.daddr == 0)
539                 goto out;
540
541         err = 0;
542         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
543                 goto out;
544
545         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
546                 t->err_count++;
547         else
548                 t->err_count = 1;
549         t->err_time = jiffies;
550 out:
551         rcu_read_unlock();
552         return err;
553 }
554
555 static inline void ipip6_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb)
556 {
557         if (INET_ECN_is_ce(iph->tos))
558                 IP6_ECN_set_ce(ipv6_hdr(skb));
559 }
560
561 static int ipip6_rcv(struct sk_buff *skb)
562 {
563         struct iphdr *iph;
564         struct ip_tunnel *tunnel;
565
566         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
567                 goto out;
568
569         iph = ip_hdr(skb);
570
571         rcu_read_lock();
572         tunnel = ipip6_tunnel_lookup(dev_net(skb->dev), skb->dev,
573                                      iph->saddr, iph->daddr);
574         if (tunnel != NULL) {
575                 struct pcpu_tstats *tstats;
576
577                 secpath_reset(skb);
578                 skb->mac_header = skb->network_header;
579                 skb_reset_network_header(skb);
580                 IPCB(skb)->flags = 0;
581                 skb->protocol = htons(ETH_P_IPV6);
582                 skb->pkt_type = PACKET_HOST;
583
584                 if ((tunnel->dev->priv_flags & IFF_ISATAP) &&
585                     !isatap_chksrc(skb, iph, tunnel)) {
586                         tunnel->dev->stats.rx_errors++;
587                         rcu_read_unlock();
588                         kfree_skb(skb);
589                         return 0;
590                 }
591
592                 tstats = this_cpu_ptr(tunnel->dev->tstats);
593                 tstats->rx_packets++;
594                 tstats->rx_bytes += skb->len;
595
596                 __skb_tunnel_rx(skb, tunnel->dev);
597
598                 ipip6_ecn_decapsulate(iph, skb);
599
600                 netif_rx(skb);
601
602                 rcu_read_unlock();
603                 return 0;
604         }
605
606         /* no tunnel matched,  let upstream know, ipsec may handle it */
607         rcu_read_unlock();
608         return 1;
609 out:
610         kfree_skb(skb);
611         return 0;
612 }
613
614 /*
615  * Returns the embedded IPv4 address if the IPv6 address
616  * comes from 6rd / 6to4 (RFC 3056) addr space.
617  */
618 static inline
619 __be32 try_6rd(struct in6_addr *v6dst, struct ip_tunnel *tunnel)
620 {
621         __be32 dst = 0;
622
623 #ifdef CONFIG_IPV6_SIT_6RD
624         if (ipv6_prefix_equal(v6dst, &tunnel->ip6rd.prefix,
625                               tunnel->ip6rd.prefixlen)) {
626                 unsigned int pbw0, pbi0;
627                 int pbi1;
628                 u32 d;
629
630                 pbw0 = tunnel->ip6rd.prefixlen >> 5;
631                 pbi0 = tunnel->ip6rd.prefixlen & 0x1f;
632
633                 d = (ntohl(v6dst->s6_addr32[pbw0]) << pbi0) >>
634                     tunnel->ip6rd.relay_prefixlen;
635
636                 pbi1 = pbi0 - tunnel->ip6rd.relay_prefixlen;
637                 if (pbi1 > 0)
638                         d |= ntohl(v6dst->s6_addr32[pbw0 + 1]) >>
639                              (32 - pbi1);
640
641                 dst = tunnel->ip6rd.relay_prefix | htonl(d);
642         }
643 #else
644         if (v6dst->s6_addr16[0] == htons(0x2002)) {
645                 /* 6to4 v6 addr has 16 bits prefix, 32 v4addr, 16 SLA, ... */
646                 memcpy(&dst, &v6dst->s6_addr16[1], 4);
647         }
648 #endif
649         return dst;
650 }
651
652 /*
653  *      This function assumes it is being called from dev_queue_xmit()
654  *      and that skb is filled properly by that function.
655  */
656
657 static netdev_tx_t ipip6_tunnel_xmit(struct sk_buff *skb,
658                                      struct net_device *dev)
659 {
660         struct ip_tunnel *tunnel = netdev_priv(dev);
661         struct pcpu_tstats *tstats;
662         struct iphdr  *tiph = &tunnel->parms.iph;
663         struct ipv6hdr *iph6 = ipv6_hdr(skb);
664         u8     tos = tunnel->parms.iph.tos;
665         __be16 df = tiph->frag_off;
666         struct rtable *rt;                      /* Route to the other host */
667         struct net_device *tdev;                /* Device to other host */
668         struct iphdr  *iph;                     /* Our new IP header */
669         unsigned int max_headroom;              /* The extra header space needed */
670         __be32 dst = tiph->daddr;
671         int    mtu;
672         struct in6_addr *addr6;
673         int addr_type;
674
675         if (skb->protocol != htons(ETH_P_IPV6))
676                 goto tx_error;
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 = skb_dst(skb)->neighbour;
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 = (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 = skb_dst(skb)->neighbour;
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 = (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), 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              =       rt->rt_dst;
825         iph->saddr              =       rt->rt_src;
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         struct iphdr *iph;
848
849         tunnel = netdev_priv(dev);
850         iph = &tunnel->parms.iph;
851
852         if (iph->daddr) {
853                 struct rtable *rt = ip_route_output_ports(dev_net(dev), NULL,
854                                                           iph->daddr, iph->saddr,
855                                                           0, 0,
856                                                           IPPROTO_IPV6,
857                                                           RT_TOS(iph->tos),
858                                                           tunnel->parms.link);
859
860                 if (!IS_ERR(rt)) {
861                         tdev = rt->dst.dev;
862                         ip_rt_put(rt);
863                 }
864                 dev->flags |= IFF_POINTOPOINT;
865         }
866
867         if (!tdev && tunnel->parms.link)
868                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
869
870         if (tdev) {
871                 dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
872                 dev->mtu = tdev->mtu - sizeof(struct iphdr);
873                 if (dev->mtu < IPV6_MIN_MTU)
874                         dev->mtu = IPV6_MIN_MTU;
875         }
876         dev->iflink = tunnel->parms.link;
877 }
878
879 static int
880 ipip6_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
881 {
882         int err = 0;
883         struct ip_tunnel_parm p;
884         struct ip_tunnel_prl prl;
885         struct ip_tunnel *t;
886         struct net *net = dev_net(dev);
887         struct sit_net *sitn = net_generic(net, sit_net_id);
888 #ifdef CONFIG_IPV6_SIT_6RD
889         struct ip_tunnel_6rd ip6rd;
890 #endif
891
892         switch (cmd) {
893         case SIOCGETTUNNEL:
894 #ifdef CONFIG_IPV6_SIT_6RD
895         case SIOCGET6RD:
896 #endif
897                 t = NULL;
898                 if (dev == sitn->fb_tunnel_dev) {
899                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
900                                 err = -EFAULT;
901                                 break;
902                         }
903                         t = ipip6_tunnel_locate(net, &p, 0);
904                 }
905                 if (t == NULL)
906                         t = netdev_priv(dev);
907
908                 err = -EFAULT;
909                 if (cmd == SIOCGETTUNNEL) {
910                         memcpy(&p, &t->parms, sizeof(p));
911                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &p,
912                                          sizeof(p)))
913                                 goto done;
914 #ifdef CONFIG_IPV6_SIT_6RD
915                 } else {
916                         ipv6_addr_copy(&ip6rd.prefix, &t->ip6rd.prefix);
917                         ip6rd.relay_prefix = t->ip6rd.relay_prefix;
918                         ip6rd.prefixlen = t->ip6rd.prefixlen;
919                         ip6rd.relay_prefixlen = t->ip6rd.relay_prefixlen;
920                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &ip6rd,
921                                          sizeof(ip6rd)))
922                                 goto done;
923 #endif
924                 }
925                 err = 0;
926                 break;
927
928         case SIOCADDTUNNEL:
929         case SIOCCHGTUNNEL:
930                 err = -EPERM;
931                 if (!capable(CAP_NET_ADMIN))
932                         goto done;
933
934                 err = -EFAULT;
935                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
936                         goto done;
937
938                 err = -EINVAL;
939                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPV6 ||
940                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
941                         goto done;
942                 if (p.iph.ttl)
943                         p.iph.frag_off |= htons(IP_DF);
944
945                 t = ipip6_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
946
947                 if (dev != sitn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
948                         if (t != NULL) {
949                                 if (t->dev != dev) {
950                                         err = -EEXIST;
951                                         break;
952                                 }
953                         } else {
954                                 if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
955                                     (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
956                                         err = -EINVAL;
957                                         break;
958                                 }
959                                 t = netdev_priv(dev);
960                                 ipip6_tunnel_unlink(sitn, t);
961                                 synchronize_net();
962                                 t->parms.iph.saddr = p.iph.saddr;
963                                 t->parms.iph.daddr = p.iph.daddr;
964                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
965                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
966                                 ipip6_tunnel_link(sitn, t);
967                                 netdev_state_change(dev);
968                         }
969                 }
970
971                 if (t) {
972                         err = 0;
973                         if (cmd == SIOCCHGTUNNEL) {
974                                 t->parms.iph.ttl = p.iph.ttl;
975                                 t->parms.iph.tos = p.iph.tos;
976                                 if (t->parms.link != p.link) {
977                                         t->parms.link = p.link;
978                                         ipip6_tunnel_bind_dev(dev);
979                                         netdev_state_change(dev);
980                                 }
981                         }
982                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
983                                 err = -EFAULT;
984                 } else
985                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
986                 break;
987
988         case SIOCDELTUNNEL:
989                 err = -EPERM;
990                 if (!capable(CAP_NET_ADMIN))
991                         goto done;
992
993                 if (dev == sitn->fb_tunnel_dev) {
994                         err = -EFAULT;
995                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
996                                 goto done;
997                         err = -ENOENT;
998                         if ((t = ipip6_tunnel_locate(net, &p, 0)) == NULL)
999                                 goto done;
1000                         err = -EPERM;
1001                         if (t == netdev_priv(sitn->fb_tunnel_dev))
1002                                 goto done;
1003                         dev = t->dev;
1004                 }
1005                 unregister_netdevice(dev);
1006                 err = 0;
1007                 break;
1008
1009         case SIOCGETPRL:
1010                 err = -EINVAL;
1011                 if (dev == sitn->fb_tunnel_dev)
1012                         goto done;
1013                 err = -ENOENT;
1014                 if (!(t = netdev_priv(dev)))
1015                         goto done;
1016                 err = ipip6_tunnel_get_prl(t, ifr->ifr_ifru.ifru_data);
1017                 break;
1018
1019         case SIOCADDPRL:
1020         case SIOCDELPRL:
1021         case SIOCCHGPRL:
1022                 err = -EPERM;
1023                 if (!capable(CAP_NET_ADMIN))
1024                         goto done;
1025                 err = -EINVAL;
1026                 if (dev == sitn->fb_tunnel_dev)
1027                         goto done;
1028                 err = -EFAULT;
1029                 if (copy_from_user(&prl, ifr->ifr_ifru.ifru_data, sizeof(prl)))
1030                         goto done;
1031                 err = -ENOENT;
1032                 if (!(t = netdev_priv(dev)))
1033                         goto done;
1034
1035                 switch (cmd) {
1036                 case SIOCDELPRL:
1037                         err = ipip6_tunnel_del_prl(t, &prl);
1038                         break;
1039                 case SIOCADDPRL:
1040                 case SIOCCHGPRL:
1041                         err = ipip6_tunnel_add_prl(t, &prl, cmd == SIOCCHGPRL);
1042                         break;
1043                 }
1044                 netdev_state_change(dev);
1045                 break;
1046
1047 #ifdef CONFIG_IPV6_SIT_6RD
1048         case SIOCADD6RD:
1049         case SIOCCHG6RD:
1050         case SIOCDEL6RD:
1051                 err = -EPERM;
1052                 if (!capable(CAP_NET_ADMIN))
1053                         goto done;
1054
1055                 err = -EFAULT;
1056                 if (copy_from_user(&ip6rd, ifr->ifr_ifru.ifru_data,
1057                                    sizeof(ip6rd)))
1058                         goto done;
1059
1060                 t = netdev_priv(dev);
1061
1062                 if (cmd != SIOCDEL6RD) {
1063                         struct in6_addr prefix;
1064                         __be32 relay_prefix;
1065
1066                         err = -EINVAL;
1067                         if (ip6rd.relay_prefixlen > 32 ||
1068                             ip6rd.prefixlen + (32 - ip6rd.relay_prefixlen) > 64)
1069                                 goto done;
1070
1071                         ipv6_addr_prefix(&prefix, &ip6rd.prefix,
1072                                          ip6rd.prefixlen);
1073                         if (!ipv6_addr_equal(&prefix, &ip6rd.prefix))
1074                                 goto done;
1075                         if (ip6rd.relay_prefixlen)
1076                                 relay_prefix = ip6rd.relay_prefix &
1077                                                htonl(0xffffffffUL <<
1078                                                      (32 - ip6rd.relay_prefixlen));
1079                         else
1080                                 relay_prefix = 0;
1081                         if (relay_prefix != ip6rd.relay_prefix)
1082                                 goto done;
1083
1084                         ipv6_addr_copy(&t->ip6rd.prefix, &prefix);
1085                         t->ip6rd.relay_prefix = relay_prefix;
1086                         t->ip6rd.prefixlen = ip6rd.prefixlen;
1087                         t->ip6rd.relay_prefixlen = ip6rd.relay_prefixlen;
1088                 } else
1089                         ipip6_tunnel_clone_6rd(dev, sitn);
1090
1091                 err = 0;
1092                 break;
1093 #endif
1094
1095         default:
1096                 err = -EINVAL;
1097         }
1098
1099 done:
1100         return err;
1101 }
1102
1103 static int ipip6_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1104 {
1105         if (new_mtu < IPV6_MIN_MTU || new_mtu > 0xFFF8 - sizeof(struct iphdr))
1106                 return -EINVAL;
1107         dev->mtu = new_mtu;
1108         return 0;
1109 }
1110
1111 static const struct net_device_ops ipip6_netdev_ops = {
1112         .ndo_uninit     = ipip6_tunnel_uninit,
1113         .ndo_start_xmit = ipip6_tunnel_xmit,
1114         .ndo_do_ioctl   = ipip6_tunnel_ioctl,
1115         .ndo_change_mtu = ipip6_tunnel_change_mtu,
1116         .ndo_get_stats  = ipip6_get_stats,
1117 };
1118
1119 static void ipip6_dev_free(struct net_device *dev)
1120 {
1121         free_percpu(dev->tstats);
1122         free_netdev(dev);
1123 }
1124
1125 static void ipip6_tunnel_setup(struct net_device *dev)
1126 {
1127         dev->netdev_ops         = &ipip6_netdev_ops;
1128         dev->destructor         = ipip6_dev_free;
1129
1130         dev->type               = ARPHRD_SIT;
1131         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr);
1132         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr);
1133         dev->flags              = IFF_NOARP;
1134         dev->priv_flags        &= ~IFF_XMIT_DST_RELEASE;
1135         dev->iflink             = 0;
1136         dev->addr_len           = 4;
1137         dev->features           |= NETIF_F_NETNS_LOCAL;
1138         dev->features           |= NETIF_F_LLTX;
1139 }
1140
1141 static int ipip6_tunnel_init(struct net_device *dev)
1142 {
1143         struct ip_tunnel *tunnel = netdev_priv(dev);
1144
1145         tunnel->dev = dev;
1146         strcpy(tunnel->parms.name, dev->name);
1147
1148         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1149         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1150
1151         ipip6_tunnel_bind_dev(dev);
1152         dev->tstats = alloc_percpu(struct pcpu_tstats);
1153         if (!dev->tstats)
1154                 return -ENOMEM;
1155
1156         return 0;
1157 }
1158
1159 static int __net_init ipip6_fb_tunnel_init(struct net_device *dev)
1160 {
1161         struct ip_tunnel *tunnel = netdev_priv(dev);
1162         struct iphdr *iph = &tunnel->parms.iph;
1163         struct net *net = dev_net(dev);
1164         struct sit_net *sitn = net_generic(net, sit_net_id);
1165
1166         tunnel->dev = dev;
1167         strcpy(tunnel->parms.name, dev->name);
1168
1169         iph->version            = 4;
1170         iph->protocol           = IPPROTO_IPV6;
1171         iph->ihl                = 5;
1172         iph->ttl                = 64;
1173
1174         dev->tstats = alloc_percpu(struct pcpu_tstats);
1175         if (!dev->tstats)
1176                 return -ENOMEM;
1177         dev_hold(dev);
1178         rcu_assign_pointer(sitn->tunnels_wc[0], tunnel);
1179         return 0;
1180 }
1181
1182 static struct xfrm_tunnel sit_handler __read_mostly = {
1183         .handler        =       ipip6_rcv,
1184         .err_handler    =       ipip6_err,
1185         .priority       =       1,
1186 };
1187
1188 static void __net_exit sit_destroy_tunnels(struct sit_net *sitn, struct list_head *head)
1189 {
1190         int prio;
1191
1192         for (prio = 1; prio < 4; prio++) {
1193                 int h;
1194                 for (h = 0; h < HASH_SIZE; h++) {
1195                         struct ip_tunnel *t;
1196
1197                         t = rtnl_dereference(sitn->tunnels[prio][h]);
1198                         while (t != NULL) {
1199                                 unregister_netdevice_queue(t->dev, head);
1200                                 t = rtnl_dereference(t->next);
1201                         }
1202                 }
1203         }
1204 }
1205
1206 static int __net_init sit_init_net(struct net *net)
1207 {
1208         struct sit_net *sitn = net_generic(net, sit_net_id);
1209         int err;
1210
1211         sitn->tunnels[0] = sitn->tunnels_wc;
1212         sitn->tunnels[1] = sitn->tunnels_l;
1213         sitn->tunnels[2] = sitn->tunnels_r;
1214         sitn->tunnels[3] = sitn->tunnels_r_l;
1215
1216         sitn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "sit0",
1217                                            ipip6_tunnel_setup);
1218         if (!sitn->fb_tunnel_dev) {
1219                 err = -ENOMEM;
1220                 goto err_alloc_dev;
1221         }
1222         dev_net_set(sitn->fb_tunnel_dev, net);
1223
1224         err = ipip6_fb_tunnel_init(sitn->fb_tunnel_dev);
1225         if (err)
1226                 goto err_dev_free;
1227
1228         ipip6_tunnel_clone_6rd(sitn->fb_tunnel_dev, sitn);
1229
1230         if ((err = register_netdev(sitn->fb_tunnel_dev)))
1231                 goto err_reg_dev;
1232
1233         return 0;
1234
1235 err_reg_dev:
1236         dev_put(sitn->fb_tunnel_dev);
1237 err_dev_free:
1238         ipip6_dev_free(sitn->fb_tunnel_dev);
1239 err_alloc_dev:
1240         return err;
1241 }
1242
1243 static void __net_exit sit_exit_net(struct net *net)
1244 {
1245         struct sit_net *sitn = net_generic(net, sit_net_id);
1246         LIST_HEAD(list);
1247
1248         rtnl_lock();
1249         sit_destroy_tunnels(sitn, &list);
1250         unregister_netdevice_queue(sitn->fb_tunnel_dev, &list);
1251         unregister_netdevice_many(&list);
1252         rtnl_unlock();
1253 }
1254
1255 static struct pernet_operations sit_net_ops = {
1256         .init = sit_init_net,
1257         .exit = sit_exit_net,
1258         .id   = &sit_net_id,
1259         .size = sizeof(struct sit_net),
1260 };
1261
1262 static void __exit sit_cleanup(void)
1263 {
1264         xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
1265
1266         unregister_pernet_device(&sit_net_ops);
1267         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1268 }
1269
1270 static int __init sit_init(void)
1271 {
1272         int err;
1273
1274         printk(KERN_INFO "IPv6 over IPv4 tunneling driver\n");
1275
1276         err = register_pernet_device(&sit_net_ops);
1277         if (err < 0)
1278                 return err;
1279         err = xfrm4_tunnel_register(&sit_handler, AF_INET6);
1280         if (err < 0) {
1281                 unregister_pernet_device(&sit_net_ops);
1282                 printk(KERN_INFO "sit init: Can't add protocol\n");
1283         }
1284         return err;
1285 }
1286
1287 module_init(sit_init);
1288 module_exit(sit_cleanup);
1289 MODULE_LICENSE("GPL");
1290 MODULE_ALIAS_NETDEV("sit0");