Merge branch 'kvm-updates-2.6.27' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6] / net / ipv4 / ipip.c
1 /*
2  *      Linux NET3:     IP/IP protocol decoder.
3  *
4  *      Authors:
5  *              Sam Lantinga (slouken@cs.ucdavis.edu)  02/01/95
6  *
7  *      Fixes:
8  *              Alan Cox        :       Merged and made usable non modular (its so tiny its silly as
9  *                                      a module taking up 2 pages).
10  *              Alan Cox        :       Fixed bug with 1.3.18 and IPIP not working (now needs to set skb->h.iph)
11  *                                      to keep ip_forward happy.
12  *              Alan Cox        :       More fixes for 1.3.21, and firewall fix. Maybe this will work soon 8).
13  *              Kai Schulte     :       Fixed #defines for IP_FIREWALL->FIREWALL
14  *              David Woodhouse :       Perform some basic ICMP handling.
15  *                                      IPIP Routing without decapsulation.
16  *              Carlos Picoto   :       GRE over IP support
17  *              Alexey Kuznetsov:       Reworked. Really, now it is truncated version of ipv4/ip_gre.c.
18  *                                      I do not want to merge them together.
19  *
20  *      This program is free software; you can redistribute it and/or
21  *      modify it under the terms of the GNU General Public License
22  *      as published by the Free Software Foundation; either version
23  *      2 of the License, or (at your option) any later version.
24  *
25  */
26
27 /* tunnel.c: an IP tunnel driver
28
29         The purpose of this driver is to provide an IP tunnel through
30         which you can tunnel network traffic transparently across subnets.
31
32         This was written by looking at Nick Holloway's dummy driver
33         Thanks for the great code!
34
35                 -Sam Lantinga   (slouken@cs.ucdavis.edu)  02/01/95
36
37         Minor tweaks:
38                 Cleaned up the code a little and added some pre-1.3.0 tweaks.
39                 dev->hard_header/hard_header_len changed to use no headers.
40                 Comments/bracketing tweaked.
41                 Made the tunnels use dev->name not tunnel: when error reporting.
42                 Added tx_dropped stat
43
44                 -Alan Cox       (Alan.Cox@linux.org) 21 March 95
45
46         Reworked:
47                 Changed to tunnel to destination gateway in addition to the
48                         tunnel's pointopoint address
49                 Almost completely rewritten
50                 Note:  There is currently no firewall or ICMP handling done.
51
52                 -Sam Lantinga   (slouken@cs.ucdavis.edu) 02/13/96
53
54 */
55
56 /* Things I wish I had known when writing the tunnel driver:
57
58         When the tunnel_xmit() function is called, the skb contains the
59         packet to be sent (plus a great deal of extra info), and dev
60         contains the tunnel device that _we_ are.
61
62         When we are passed a packet, we are expected to fill in the
63         source address with our source IP address.
64
65         What is the proper way to allocate, copy and free a buffer?
66         After you allocate it, it is a "0 length" chunk of memory
67         starting at zero.  If you want to add headers to the buffer
68         later, you'll have to call "skb_reserve(skb, amount)" with
69         the amount of memory you want reserved.  Then, you call
70         "skb_put(skb, amount)" with the amount of space you want in
71         the buffer.  skb_put() returns a pointer to the top (#0) of
72         that buffer.  skb->len is set to the amount of space you have
73         "allocated" with skb_put().  You can then write up to skb->len
74         bytes to that buffer.  If you need more, you can call skb_put()
75         again with the additional amount of space you need.  You can
76         find out how much more space you can allocate by calling
77         "skb_tailroom(skb)".
78         Now, to add header space, call "skb_push(skb, header_len)".
79         This creates space at the beginning of the buffer and returns
80         a pointer to this new space.  If later you need to strip a
81         header from a buffer, call "skb_pull(skb, header_len)".
82         skb_headroom() will return how much space is left at the top
83         of the buffer (before the main data).  Remember, this headroom
84         space must be reserved before the skb_put() function is called.
85         */
86
87 /*
88    This version of net/ipv4/ipip.c is cloned of net/ipv4/ip_gre.c
89
90    For comments look at net/ipv4/ip_gre.c --ANK
91  */
92
93
94 #include <linux/capability.h>
95 #include <linux/module.h>
96 #include <linux/types.h>
97 #include <linux/kernel.h>
98 #include <asm/uaccess.h>
99 #include <linux/skbuff.h>
100 #include <linux/netdevice.h>
101 #include <linux/in.h>
102 #include <linux/tcp.h>
103 #include <linux/udp.h>
104 #include <linux/if_arp.h>
105 #include <linux/mroute.h>
106 #include <linux/init.h>
107 #include <linux/netfilter_ipv4.h>
108 #include <linux/if_ether.h>
109
110 #include <net/sock.h>
111 #include <net/ip.h>
112 #include <net/icmp.h>
113 #include <net/ipip.h>
114 #include <net/inet_ecn.h>
115 #include <net/xfrm.h>
116 #include <net/net_namespace.h>
117 #include <net/netns/generic.h>
118
119 #define HASH_SIZE  16
120 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
121
122 static int ipip_net_id;
123 struct ipip_net {
124         struct ip_tunnel *tunnels_r_l[HASH_SIZE];
125         struct ip_tunnel *tunnels_r[HASH_SIZE];
126         struct ip_tunnel *tunnels_l[HASH_SIZE];
127         struct ip_tunnel *tunnels_wc[1];
128         struct ip_tunnel **tunnels[4];
129
130         struct net_device *fb_tunnel_dev;
131 };
132
133 static int ipip_fb_tunnel_init(struct net_device *dev);
134 static int ipip_tunnel_init(struct net_device *dev);
135 static void ipip_tunnel_setup(struct net_device *dev);
136
137 static DEFINE_RWLOCK(ipip_lock);
138
139 static struct ip_tunnel * ipip_tunnel_lookup(struct net *net,
140                 __be32 remote, __be32 local)
141 {
142         unsigned h0 = HASH(remote);
143         unsigned h1 = HASH(local);
144         struct ip_tunnel *t;
145         struct ipip_net *ipn = net_generic(net, ipip_net_id);
146
147         for (t = ipn->tunnels_r_l[h0^h1]; t; t = t->next) {
148                 if (local == t->parms.iph.saddr &&
149                     remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
150                         return t;
151         }
152         for (t = ipn->tunnels_r[h0]; t; t = t->next) {
153                 if (remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
154                         return t;
155         }
156         for (t = ipn->tunnels_l[h1]; t; t = t->next) {
157                 if (local == t->parms.iph.saddr && (t->dev->flags&IFF_UP))
158                         return t;
159         }
160         if ((t = ipn->tunnels_wc[0]) != NULL && (t->dev->flags&IFF_UP))
161                 return t;
162         return NULL;
163 }
164
165 static struct ip_tunnel **__ipip_bucket(struct ipip_net *ipn,
166                 struct ip_tunnel_parm *parms)
167 {
168         __be32 remote = parms->iph.daddr;
169         __be32 local = parms->iph.saddr;
170         unsigned h = 0;
171         int prio = 0;
172
173         if (remote) {
174                 prio |= 2;
175                 h ^= HASH(remote);
176         }
177         if (local) {
178                 prio |= 1;
179                 h ^= HASH(local);
180         }
181         return &ipn->tunnels[prio][h];
182 }
183
184 static inline struct ip_tunnel **ipip_bucket(struct ipip_net *ipn,
185                 struct ip_tunnel *t)
186 {
187         return __ipip_bucket(ipn, &t->parms);
188 }
189
190 static void ipip_tunnel_unlink(struct ipip_net *ipn, struct ip_tunnel *t)
191 {
192         struct ip_tunnel **tp;
193
194         for (tp = ipip_bucket(ipn, t); *tp; tp = &(*tp)->next) {
195                 if (t == *tp) {
196                         write_lock_bh(&ipip_lock);
197                         *tp = t->next;
198                         write_unlock_bh(&ipip_lock);
199                         break;
200                 }
201         }
202 }
203
204 static void ipip_tunnel_link(struct ipip_net *ipn, struct ip_tunnel *t)
205 {
206         struct ip_tunnel **tp = ipip_bucket(ipn, t);
207
208         t->next = *tp;
209         write_lock_bh(&ipip_lock);
210         *tp = t;
211         write_unlock_bh(&ipip_lock);
212 }
213
214 static struct ip_tunnel * ipip_tunnel_locate(struct net *net,
215                 struct ip_tunnel_parm *parms, int create)
216 {
217         __be32 remote = parms->iph.daddr;
218         __be32 local = parms->iph.saddr;
219         struct ip_tunnel *t, **tp, *nt;
220         struct net_device *dev;
221         char name[IFNAMSIZ];
222         struct ipip_net *ipn = net_generic(net, ipip_net_id);
223
224         for (tp = __ipip_bucket(ipn, parms); (t = *tp) != NULL; tp = &t->next) {
225                 if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr)
226                         return t;
227         }
228         if (!create)
229                 return NULL;
230
231         if (parms->name[0])
232                 strlcpy(name, parms->name, IFNAMSIZ);
233         else
234                 sprintf(name, "tunl%%d");
235
236         dev = alloc_netdev(sizeof(*t), name, ipip_tunnel_setup);
237         if (dev == NULL)
238                 return NULL;
239
240         dev_net_set(dev, net);
241
242         if (strchr(name, '%')) {
243                 if (dev_alloc_name(dev, name) < 0)
244                         goto failed_free;
245         }
246
247         nt = netdev_priv(dev);
248         dev->init = ipip_tunnel_init;
249         nt->parms = *parms;
250
251         if (register_netdevice(dev) < 0)
252                 goto failed_free;
253
254         dev_hold(dev);
255         ipip_tunnel_link(ipn, nt);
256         return nt;
257
258 failed_free:
259         free_netdev(dev);
260         return NULL;
261 }
262
263 static void ipip_tunnel_uninit(struct net_device *dev)
264 {
265         struct net *net = dev_net(dev);
266         struct ipip_net *ipn = net_generic(net, ipip_net_id);
267
268         if (dev == ipn->fb_tunnel_dev) {
269                 write_lock_bh(&ipip_lock);
270                 ipn->tunnels_wc[0] = NULL;
271                 write_unlock_bh(&ipip_lock);
272         } else
273                 ipip_tunnel_unlink(ipn, netdev_priv(dev));
274         dev_put(dev);
275 }
276
277 static int ipip_err(struct sk_buff *skb, u32 info)
278 {
279
280 /* All the routers (except for Linux) return only
281    8 bytes of packet payload. It means, that precise relaying of
282    ICMP in the real Internet is absolutely infeasible.
283  */
284         struct iphdr *iph = (struct iphdr*)skb->data;
285         const int type = icmp_hdr(skb)->type;
286         const int code = icmp_hdr(skb)->code;
287         struct ip_tunnel *t;
288         int err;
289
290         switch (type) {
291         default:
292         case ICMP_PARAMETERPROB:
293                 return 0;
294
295         case ICMP_DEST_UNREACH:
296                 switch (code) {
297                 case ICMP_SR_FAILED:
298                 case ICMP_PORT_UNREACH:
299                         /* Impossible event. */
300                         return 0;
301                 case ICMP_FRAG_NEEDED:
302                         /* Soft state for pmtu is maintained by IP core. */
303                         return 0;
304                 default:
305                         /* All others are translated to HOST_UNREACH.
306                            rfc2003 contains "deep thoughts" about NET_UNREACH,
307                            I believe they are just ether pollution. --ANK
308                          */
309                         break;
310                 }
311                 break;
312         case ICMP_TIME_EXCEEDED:
313                 if (code != ICMP_EXC_TTL)
314                         return 0;
315                 break;
316         }
317
318         err = -ENOENT;
319
320         read_lock(&ipip_lock);
321         t = ipip_tunnel_lookup(dev_net(skb->dev), iph->daddr, iph->saddr);
322         if (t == NULL || t->parms.iph.daddr == 0)
323                 goto out;
324
325         err = 0;
326         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
327                 goto out;
328
329         if (jiffies - t->err_time < IPTUNNEL_ERR_TIMEO)
330                 t->err_count++;
331         else
332                 t->err_count = 1;
333         t->err_time = jiffies;
334 out:
335         read_unlock(&ipip_lock);
336         return err;
337 }
338
339 static inline void ipip_ecn_decapsulate(const struct iphdr *outer_iph,
340                                         struct sk_buff *skb)
341 {
342         struct iphdr *inner_iph = ip_hdr(skb);
343
344         if (INET_ECN_is_ce(outer_iph->tos))
345                 IP_ECN_set_ce(inner_iph);
346 }
347
348 static int ipip_rcv(struct sk_buff *skb)
349 {
350         struct ip_tunnel *tunnel;
351         const struct iphdr *iph = ip_hdr(skb);
352
353         read_lock(&ipip_lock);
354         if ((tunnel = ipip_tunnel_lookup(dev_net(skb->dev),
355                                         iph->saddr, iph->daddr)) != NULL) {
356                 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
357                         read_unlock(&ipip_lock);
358                         kfree_skb(skb);
359                         return 0;
360                 }
361
362                 secpath_reset(skb);
363
364                 skb->mac_header = skb->network_header;
365                 skb_reset_network_header(skb);
366                 skb->protocol = htons(ETH_P_IP);
367                 skb->pkt_type = PACKET_HOST;
368
369                 tunnel->dev->stats.rx_packets++;
370                 tunnel->dev->stats.rx_bytes += skb->len;
371                 skb->dev = tunnel->dev;
372                 dst_release(skb->dst);
373                 skb->dst = NULL;
374                 nf_reset(skb);
375                 ipip_ecn_decapsulate(iph, skb);
376                 netif_rx(skb);
377                 read_unlock(&ipip_lock);
378                 return 0;
379         }
380         read_unlock(&ipip_lock);
381
382         return -1;
383 }
384
385 /*
386  *      This function assumes it is being called from dev_queue_xmit()
387  *      and that skb is filled properly by that function.
388  */
389
390 static int ipip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
391 {
392         struct ip_tunnel *tunnel = netdev_priv(dev);
393         struct net_device_stats *stats = &tunnel->dev->stats;
394         struct iphdr  *tiph = &tunnel->parms.iph;
395         u8     tos = tunnel->parms.iph.tos;
396         __be16 df = tiph->frag_off;
397         struct rtable *rt;                      /* Route to the other host */
398         struct net_device *tdev;                        /* Device to other host */
399         struct iphdr  *old_iph = ip_hdr(skb);
400         struct iphdr  *iph;                     /* Our new IP header */
401         unsigned int max_headroom;              /* The extra header space needed */
402         __be32 dst = tiph->daddr;
403         int    mtu;
404
405         if (tunnel->recursion++) {
406                 stats->collisions++;
407                 goto tx_error;
408         }
409
410         if (skb->protocol != htons(ETH_P_IP))
411                 goto tx_error;
412
413         if (tos&1)
414                 tos = old_iph->tos;
415
416         if (!dst) {
417                 /* NBMA tunnel */
418                 if ((rt = skb->rtable) == NULL) {
419                         stats->tx_fifo_errors++;
420                         goto tx_error;
421                 }
422                 if ((dst = rt->rt_gateway) == 0)
423                         goto tx_error_icmp;
424         }
425
426         {
427                 struct flowi fl = { .oif = tunnel->parms.link,
428                                     .nl_u = { .ip4_u =
429                                               { .daddr = dst,
430                                                 .saddr = tiph->saddr,
431                                                 .tos = RT_TOS(tos) } },
432                                     .proto = IPPROTO_IPIP };
433                 if (ip_route_output_key(dev_net(dev), &rt, &fl)) {
434                         stats->tx_carrier_errors++;
435                         goto tx_error_icmp;
436                 }
437         }
438         tdev = rt->u.dst.dev;
439
440         if (tdev == dev) {
441                 ip_rt_put(rt);
442                 stats->collisions++;
443                 goto tx_error;
444         }
445
446         if (tiph->frag_off)
447                 mtu = dst_mtu(&rt->u.dst) - sizeof(struct iphdr);
448         else
449                 mtu = skb->dst ? dst_mtu(skb->dst) : dev->mtu;
450
451         if (mtu < 68) {
452                 stats->collisions++;
453                 ip_rt_put(rt);
454                 goto tx_error;
455         }
456         if (skb->dst)
457                 skb->dst->ops->update_pmtu(skb->dst, mtu);
458
459         df |= (old_iph->frag_off&htons(IP_DF));
460
461         if ((old_iph->frag_off&htons(IP_DF)) && mtu < ntohs(old_iph->tot_len)) {
462                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
463                 ip_rt_put(rt);
464                 goto tx_error;
465         }
466
467         if (tunnel->err_count > 0) {
468                 if (jiffies - tunnel->err_time < IPTUNNEL_ERR_TIMEO) {
469                         tunnel->err_count--;
470                         dst_link_failure(skb);
471                 } else
472                         tunnel->err_count = 0;
473         }
474
475         /*
476          * Okay, now see if we can stuff it in the buffer as-is.
477          */
478         max_headroom = (LL_RESERVED_SPACE(tdev)+sizeof(struct iphdr));
479
480         if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
481             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
482                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
483                 if (!new_skb) {
484                         ip_rt_put(rt);
485                         stats->tx_dropped++;
486                         dev_kfree_skb(skb);
487                         tunnel->recursion--;
488                         return 0;
489                 }
490                 if (skb->sk)
491                         skb_set_owner_w(new_skb, skb->sk);
492                 dev_kfree_skb(skb);
493                 skb = new_skb;
494                 old_iph = ip_hdr(skb);
495         }
496
497         skb->transport_header = skb->network_header;
498         skb_push(skb, sizeof(struct iphdr));
499         skb_reset_network_header(skb);
500         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
501         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
502                               IPSKB_REROUTED);
503         dst_release(skb->dst);
504         skb->dst = &rt->u.dst;
505
506         /*
507          *      Push down and install the IPIP header.
508          */
509
510         iph                     =       ip_hdr(skb);
511         iph->version            =       4;
512         iph->ihl                =       sizeof(struct iphdr)>>2;
513         iph->frag_off           =       df;
514         iph->protocol           =       IPPROTO_IPIP;
515         iph->tos                =       INET_ECN_encapsulate(tos, old_iph->tos);
516         iph->daddr              =       rt->rt_dst;
517         iph->saddr              =       rt->rt_src;
518
519         if ((iph->ttl = tiph->ttl) == 0)
520                 iph->ttl        =       old_iph->ttl;
521
522         nf_reset(skb);
523
524         IPTUNNEL_XMIT();
525         tunnel->recursion--;
526         return 0;
527
528 tx_error_icmp:
529         dst_link_failure(skb);
530 tx_error:
531         stats->tx_errors++;
532         dev_kfree_skb(skb);
533         tunnel->recursion--;
534         return 0;
535 }
536
537 static void ipip_tunnel_bind_dev(struct net_device *dev)
538 {
539         struct net_device *tdev = NULL;
540         struct ip_tunnel *tunnel;
541         struct iphdr *iph;
542
543         tunnel = netdev_priv(dev);
544         iph = &tunnel->parms.iph;
545
546         if (iph->daddr) {
547                 struct flowi fl = { .oif = tunnel->parms.link,
548                                     .nl_u = { .ip4_u =
549                                               { .daddr = iph->daddr,
550                                                 .saddr = iph->saddr,
551                                                 .tos = RT_TOS(iph->tos) } },
552                                     .proto = IPPROTO_IPIP };
553                 struct rtable *rt;
554                 if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
555                         tdev = rt->u.dst.dev;
556                         ip_rt_put(rt);
557                 }
558                 dev->flags |= IFF_POINTOPOINT;
559         }
560
561         if (!tdev && tunnel->parms.link)
562                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
563
564         if (tdev) {
565                 dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
566                 dev->mtu = tdev->mtu - sizeof(struct iphdr);
567         }
568         dev->iflink = tunnel->parms.link;
569 }
570
571 static int
572 ipip_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
573 {
574         int err = 0;
575         struct ip_tunnel_parm p;
576         struct ip_tunnel *t;
577         struct net *net = dev_net(dev);
578         struct ipip_net *ipn = net_generic(net, ipip_net_id);
579
580         switch (cmd) {
581         case SIOCGETTUNNEL:
582                 t = NULL;
583                 if (dev == ipn->fb_tunnel_dev) {
584                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
585                                 err = -EFAULT;
586                                 break;
587                         }
588                         t = ipip_tunnel_locate(net, &p, 0);
589                 }
590                 if (t == NULL)
591                         t = netdev_priv(dev);
592                 memcpy(&p, &t->parms, sizeof(p));
593                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
594                         err = -EFAULT;
595                 break;
596
597         case SIOCADDTUNNEL:
598         case SIOCCHGTUNNEL:
599                 err = -EPERM;
600                 if (!capable(CAP_NET_ADMIN))
601                         goto done;
602
603                 err = -EFAULT;
604                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
605                         goto done;
606
607                 err = -EINVAL;
608                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPIP ||
609                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
610                         goto done;
611                 if (p.iph.ttl)
612                         p.iph.frag_off |= htons(IP_DF);
613
614                 t = ipip_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
615
616                 if (dev != ipn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
617                         if (t != NULL) {
618                                 if (t->dev != dev) {
619                                         err = -EEXIST;
620                                         break;
621                                 }
622                         } else {
623                                 if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
624                                     (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
625                                         err = -EINVAL;
626                                         break;
627                                 }
628                                 t = netdev_priv(dev);
629                                 ipip_tunnel_unlink(ipn, t);
630                                 t->parms.iph.saddr = p.iph.saddr;
631                                 t->parms.iph.daddr = p.iph.daddr;
632                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
633                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
634                                 ipip_tunnel_link(ipn, t);
635                                 netdev_state_change(dev);
636                         }
637                 }
638
639                 if (t) {
640                         err = 0;
641                         if (cmd == SIOCCHGTUNNEL) {
642                                 t->parms.iph.ttl = p.iph.ttl;
643                                 t->parms.iph.tos = p.iph.tos;
644                                 t->parms.iph.frag_off = p.iph.frag_off;
645                                 if (t->parms.link != p.link) {
646                                         t->parms.link = p.link;
647                                         ipip_tunnel_bind_dev(dev);
648                                         netdev_state_change(dev);
649                                 }
650                         }
651                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
652                                 err = -EFAULT;
653                 } else
654                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
655                 break;
656
657         case SIOCDELTUNNEL:
658                 err = -EPERM;
659                 if (!capable(CAP_NET_ADMIN))
660                         goto done;
661
662                 if (dev == ipn->fb_tunnel_dev) {
663                         err = -EFAULT;
664                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
665                                 goto done;
666                         err = -ENOENT;
667                         if ((t = ipip_tunnel_locate(net, &p, 0)) == NULL)
668                                 goto done;
669                         err = -EPERM;
670                         if (t->dev == ipn->fb_tunnel_dev)
671                                 goto done;
672                         dev = t->dev;
673                 }
674                 unregister_netdevice(dev);
675                 err = 0;
676                 break;
677
678         default:
679                 err = -EINVAL;
680         }
681
682 done:
683         return err;
684 }
685
686 static int ipip_tunnel_change_mtu(struct net_device *dev, int new_mtu)
687 {
688         if (new_mtu < 68 || new_mtu > 0xFFF8 - sizeof(struct iphdr))
689                 return -EINVAL;
690         dev->mtu = new_mtu;
691         return 0;
692 }
693
694 static void ipip_tunnel_setup(struct net_device *dev)
695 {
696         dev->uninit             = ipip_tunnel_uninit;
697         dev->hard_start_xmit    = ipip_tunnel_xmit;
698         dev->do_ioctl           = ipip_tunnel_ioctl;
699         dev->change_mtu         = ipip_tunnel_change_mtu;
700         dev->destructor         = free_netdev;
701
702         dev->type               = ARPHRD_TUNNEL;
703         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr);
704         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr);
705         dev->flags              = IFF_NOARP;
706         dev->iflink             = 0;
707         dev->addr_len           = 4;
708         dev->features           |= NETIF_F_NETNS_LOCAL;
709 }
710
711 static int ipip_tunnel_init(struct net_device *dev)
712 {
713         struct ip_tunnel *tunnel;
714
715         tunnel = netdev_priv(dev);
716
717         tunnel->dev = dev;
718         strcpy(tunnel->parms.name, dev->name);
719
720         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
721         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
722
723         ipip_tunnel_bind_dev(dev);
724
725         return 0;
726 }
727
728 static int ipip_fb_tunnel_init(struct net_device *dev)
729 {
730         struct ip_tunnel *tunnel = netdev_priv(dev);
731         struct iphdr *iph = &tunnel->parms.iph;
732         struct ipip_net *ipn = net_generic(dev_net(dev), ipip_net_id);
733
734         tunnel->dev = dev;
735         strcpy(tunnel->parms.name, dev->name);
736
737         iph->version            = 4;
738         iph->protocol           = IPPROTO_IPIP;
739         iph->ihl                = 5;
740
741         dev_hold(dev);
742         ipn->tunnels_wc[0]      = tunnel;
743         return 0;
744 }
745
746 static struct xfrm_tunnel ipip_handler = {
747         .handler        =       ipip_rcv,
748         .err_handler    =       ipip_err,
749         .priority       =       1,
750 };
751
752 static char banner[] __initdata =
753         KERN_INFO "IPv4 over IPv4 tunneling driver\n";
754
755 static void ipip_destroy_tunnels(struct ipip_net *ipn)
756 {
757         int prio;
758
759         for (prio = 1; prio < 4; prio++) {
760                 int h;
761                 for (h = 0; h < HASH_SIZE; h++) {
762                         struct ip_tunnel *t;
763                         while ((t = ipn->tunnels[prio][h]) != NULL)
764                                 unregister_netdevice(t->dev);
765                 }
766         }
767 }
768
769 static int ipip_init_net(struct net *net)
770 {
771         int err;
772         struct ipip_net *ipn;
773
774         err = -ENOMEM;
775         ipn = kzalloc(sizeof(struct ipip_net), GFP_KERNEL);
776         if (ipn == NULL)
777                 goto err_alloc;
778
779         err = net_assign_generic(net, ipip_net_id, ipn);
780         if (err < 0)
781                 goto err_assign;
782
783         ipn->tunnels[0] = ipn->tunnels_wc;
784         ipn->tunnels[1] = ipn->tunnels_l;
785         ipn->tunnels[2] = ipn->tunnels_r;
786         ipn->tunnels[3] = ipn->tunnels_r_l;
787
788         ipn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel),
789                                            "tunl0",
790                                            ipip_tunnel_setup);
791         if (!ipn->fb_tunnel_dev) {
792                 err = -ENOMEM;
793                 goto err_alloc_dev;
794         }
795
796         ipn->fb_tunnel_dev->init = ipip_fb_tunnel_init;
797         dev_net_set(ipn->fb_tunnel_dev, net);
798
799         if ((err = register_netdev(ipn->fb_tunnel_dev)))
800                 goto err_reg_dev;
801
802         return 0;
803
804 err_reg_dev:
805         free_netdev(ipn->fb_tunnel_dev);
806 err_alloc_dev:
807         /* nothing */
808 err_assign:
809         kfree(ipn);
810 err_alloc:
811         return err;
812 }
813
814 static void ipip_exit_net(struct net *net)
815 {
816         struct ipip_net *ipn;
817
818         ipn = net_generic(net, ipip_net_id);
819         rtnl_lock();
820         ipip_destroy_tunnels(ipn);
821         unregister_netdevice(ipn->fb_tunnel_dev);
822         rtnl_unlock();
823         kfree(ipn);
824 }
825
826 static struct pernet_operations ipip_net_ops = {
827         .init = ipip_init_net,
828         .exit = ipip_exit_net,
829 };
830
831 static int __init ipip_init(void)
832 {
833         int err;
834
835         printk(banner);
836
837         if (xfrm4_tunnel_register(&ipip_handler, AF_INET)) {
838                 printk(KERN_INFO "ipip init: can't register tunnel\n");
839                 return -EAGAIN;
840         }
841
842         err = register_pernet_gen_device(&ipip_net_id, &ipip_net_ops);
843         if (err)
844                 xfrm4_tunnel_deregister(&ipip_handler, AF_INET);
845
846         return err;
847 }
848
849 static void __exit ipip_fini(void)
850 {
851         if (xfrm4_tunnel_deregister(&ipip_handler, AF_INET))
852                 printk(KERN_INFO "ipip close: can't deregister tunnel\n");
853
854         unregister_pernet_gen_device(ipip_net_id, &ipip_net_ops);
855 }
856
857 module_init(ipip_init);
858 module_exit(ipip_fini);
859 MODULE_LICENSE("GPL");