Merge with /pub/scm/linux/kernel/git/torvalds/linux-2.6.git
[linux-2.6] / net / ipv4 / ipip.c
1 /*
2  *      Linux NET3:     IP/IP protocol decoder. 
3  *
4  *      Version: $Id: ipip.c,v 1.50 2001/10/02 02:22:36 davem Exp $
5  *
6  *      Authors:
7  *              Sam Lantinga (slouken@cs.ucdavis.edu)  02/01/95
8  *
9  *      Fixes:
10  *              Alan Cox        :       Merged and made usable non modular (its so tiny its silly as
11  *                                      a module taking up 2 pages).
12  *              Alan Cox        :       Fixed bug with 1.3.18 and IPIP not working (now needs to set skb->h.iph)
13  *                                      to keep ip_forward happy.
14  *              Alan Cox        :       More fixes for 1.3.21, and firewall fix. Maybe this will work soon 8).
15  *              Kai Schulte     :       Fixed #defines for IP_FIREWALL->FIREWALL
16  *              David Woodhouse :       Perform some basic ICMP handling.
17  *                                      IPIP Routing without decapsulation.
18  *              Carlos Picoto   :       GRE over IP support
19  *              Alexey Kuznetsov:       Reworked. Really, now it is truncated version of ipv4/ip_gre.c.
20  *                                      I do not want to merge them together.
21  *
22  *      This program is free software; you can redistribute it and/or
23  *      modify it under the terms of the GNU General Public License
24  *      as published by the Free Software Foundation; either version
25  *      2 of the License, or (at your option) any later version.
26  *
27  */
28
29 /* tunnel.c: an IP tunnel driver
30
31         The purpose of this driver is to provide an IP tunnel through
32         which you can tunnel network traffic transparently across subnets.
33
34         This was written by looking at Nick Holloway's dummy driver
35         Thanks for the great code!
36
37                 -Sam Lantinga   (slouken@cs.ucdavis.edu)  02/01/95
38                 
39         Minor tweaks:
40                 Cleaned up the code a little and added some pre-1.3.0 tweaks.
41                 dev->hard_header/hard_header_len changed to use no headers.
42                 Comments/bracketing tweaked.
43                 Made the tunnels use dev->name not tunnel: when error reporting.
44                 Added tx_dropped stat
45                 
46                 -Alan Cox       (Alan.Cox@linux.org) 21 March 95
47
48         Reworked:
49                 Changed to tunnel to destination gateway in addition to the
50                         tunnel's pointopoint address
51                 Almost completely rewritten
52                 Note:  There is currently no firewall or ICMP handling done.
53
54                 -Sam Lantinga   (slouken@cs.ucdavis.edu) 02/13/96
55                 
56 */
57
58 /* Things I wish I had known when writing the tunnel driver:
59
60         When the tunnel_xmit() function is called, the skb contains the
61         packet to be sent (plus a great deal of extra info), and dev
62         contains the tunnel device that _we_ are.
63
64         When we are passed a packet, we are expected to fill in the
65         source address with our source IP address.
66
67         What is the proper way to allocate, copy and free a buffer?
68         After you allocate it, it is a "0 length" chunk of memory
69         starting at zero.  If you want to add headers to the buffer
70         later, you'll have to call "skb_reserve(skb, amount)" with
71         the amount of memory you want reserved.  Then, you call
72         "skb_put(skb, amount)" with the amount of space you want in
73         the buffer.  skb_put() returns a pointer to the top (#0) of
74         that buffer.  skb->len is set to the amount of space you have
75         "allocated" with skb_put().  You can then write up to skb->len
76         bytes to that buffer.  If you need more, you can call skb_put()
77         again with the additional amount of space you need.  You can
78         find out how much more space you can allocate by calling 
79         "skb_tailroom(skb)".
80         Now, to add header space, call "skb_push(skb, header_len)".
81         This creates space at the beginning of the buffer and returns
82         a pointer to this new space.  If later you need to strip a
83         header from a buffer, call "skb_pull(skb, header_len)".
84         skb_headroom() will return how much space is left at the top
85         of the buffer (before the main data).  Remember, this headroom
86         space must be reserved before the skb_put() function is called.
87         */
88
89 /*
90    This version of net/ipv4/ipip.c is cloned of net/ipv4/ip_gre.c
91
92    For comments look at net/ipv4/ip_gre.c --ANK
93  */
94
95  
96 #include <linux/capability.h>
97 #include <linux/config.h>
98 #include <linux/module.h>
99 #include <linux/types.h>
100 #include <linux/sched.h>
101 #include <linux/kernel.h>
102 #include <asm/uaccess.h>
103 #include <linux/skbuff.h>
104 #include <linux/netdevice.h>
105 #include <linux/in.h>
106 #include <linux/tcp.h>
107 #include <linux/udp.h>
108 #include <linux/if_arp.h>
109 #include <linux/mroute.h>
110 #include <linux/init.h>
111 #include <linux/netfilter_ipv4.h>
112 #include <linux/if_ether.h>
113
114 #include <net/sock.h>
115 #include <net/ip.h>
116 #include <net/icmp.h>
117 #include <net/ipip.h>
118 #include <net/inet_ecn.h>
119 #include <net/xfrm.h>
120
121 #define HASH_SIZE  16
122 #define HASH(addr) ((addr^(addr>>4))&0xF)
123
124 static int ipip_fb_tunnel_init(struct net_device *dev);
125 static int ipip_tunnel_init(struct net_device *dev);
126 static void ipip_tunnel_setup(struct net_device *dev);
127
128 static struct net_device *ipip_fb_tunnel_dev;
129
130 static struct ip_tunnel *tunnels_r_l[HASH_SIZE];
131 static struct ip_tunnel *tunnels_r[HASH_SIZE];
132 static struct ip_tunnel *tunnels_l[HASH_SIZE];
133 static struct ip_tunnel *tunnels_wc[1];
134 static struct ip_tunnel **tunnels[4] = { tunnels_wc, tunnels_l, tunnels_r, tunnels_r_l };
135
136 static DEFINE_RWLOCK(ipip_lock);
137
138 static struct ip_tunnel * ipip_tunnel_lookup(u32 remote, u32 local)
139 {
140         unsigned h0 = HASH(remote);
141         unsigned h1 = HASH(local);
142         struct ip_tunnel *t;
143
144         for (t = tunnels_r_l[h0^h1]; t; t = t->next) {
145                 if (local == t->parms.iph.saddr &&
146                     remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
147                         return t;
148         }
149         for (t = tunnels_r[h0]; t; t = t->next) {
150                 if (remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
151                         return t;
152         }
153         for (t = tunnels_l[h1]; t; t = t->next) {
154                 if (local == t->parms.iph.saddr && (t->dev->flags&IFF_UP))
155                         return t;
156         }
157         if ((t = tunnels_wc[0]) != NULL && (t->dev->flags&IFF_UP))
158                 return t;
159         return NULL;
160 }
161
162 static struct ip_tunnel **ipip_bucket(struct ip_tunnel *t)
163 {
164         u32 remote = t->parms.iph.daddr;
165         u32 local = t->parms.iph.saddr;
166         unsigned h = 0;
167         int prio = 0;
168
169         if (remote) {
170                 prio |= 2;
171                 h ^= HASH(remote);
172         }
173         if (local) {
174                 prio |= 1;
175                 h ^= HASH(local);
176         }
177         return &tunnels[prio][h];
178 }
179
180
181 static void ipip_tunnel_unlink(struct ip_tunnel *t)
182 {
183         struct ip_tunnel **tp;
184
185         for (tp = ipip_bucket(t); *tp; tp = &(*tp)->next) {
186                 if (t == *tp) {
187                         write_lock_bh(&ipip_lock);
188                         *tp = t->next;
189                         write_unlock_bh(&ipip_lock);
190                         break;
191                 }
192         }
193 }
194
195 static void ipip_tunnel_link(struct ip_tunnel *t)
196 {
197         struct ip_tunnel **tp = ipip_bucket(t);
198
199         t->next = *tp;
200         write_lock_bh(&ipip_lock);
201         *tp = t;
202         write_unlock_bh(&ipip_lock);
203 }
204
205 static struct ip_tunnel * ipip_tunnel_locate(struct ip_tunnel_parm *parms, int create)
206 {
207         u32 remote = parms->iph.daddr;
208         u32 local = parms->iph.saddr;
209         struct ip_tunnel *t, **tp, *nt;
210         struct net_device *dev;
211         unsigned h = 0;
212         int prio = 0;
213         char name[IFNAMSIZ];
214
215         if (remote) {
216                 prio |= 2;
217                 h ^= HASH(remote);
218         }
219         if (local) {
220                 prio |= 1;
221                 h ^= HASH(local);
222         }
223         for (tp = &tunnels[prio][h]; (t = *tp) != NULL; tp = &t->next) {
224                 if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr)
225                         return t;
226         }
227         if (!create)
228                 return NULL;
229
230         if (parms->name[0])
231                 strlcpy(name, parms->name, IFNAMSIZ);
232         else {
233                 int i;
234                 for (i=1; i<100; i++) {
235                         sprintf(name, "tunl%d", i);
236                         if (__dev_get_by_name(name) == NULL)
237                                 break;
238                 }
239                 if (i==100)
240                         goto failed;
241         }
242
243         dev = alloc_netdev(sizeof(*t), name, ipip_tunnel_setup);
244         if (dev == NULL)
245                 return NULL;
246
247         nt = netdev_priv(dev);
248         SET_MODULE_OWNER(dev);
249         dev->init = ipip_tunnel_init;
250         nt->parms = *parms;
251
252         if (register_netdevice(dev) < 0) {
253                 free_netdev(dev);
254                 goto failed;
255         }
256
257         dev_hold(dev);
258         ipip_tunnel_link(nt);
259         return nt;
260
261 failed:
262         return NULL;
263 }
264
265 static void ipip_tunnel_uninit(struct net_device *dev)
266 {
267         if (dev == ipip_fb_tunnel_dev) {
268                 write_lock_bh(&ipip_lock);
269                 tunnels_wc[0] = NULL;
270                 write_unlock_bh(&ipip_lock);
271         } else
272                 ipip_tunnel_unlink(netdev_priv(dev));
273         dev_put(dev);
274 }
275
276 static int ipip_err(struct sk_buff *skb, u32 info)
277 {
278 #ifndef I_WISH_WORLD_WERE_PERFECT
279
280 /* It is not :-( 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         int type = skb->h.icmph->type;
286         int code = skb->h.icmph->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(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 #else
338         struct iphdr *iph = (struct iphdr*)dp;
339         int hlen = iph->ihl<<2;
340         struct iphdr *eiph;
341         int type = skb->h.icmph->type;
342         int code = skb->h.icmph->code;
343         int rel_type = 0;
344         int rel_code = 0;
345         int rel_info = 0;
346         struct sk_buff *skb2;
347         struct flowi fl;
348         struct rtable *rt;
349
350         if (len < hlen + sizeof(struct iphdr))
351                 return 0;
352         eiph = (struct iphdr*)(dp + hlen);
353
354         switch (type) {
355         default:
356                 return 0;
357         case ICMP_PARAMETERPROB:
358                 if (skb->h.icmph->un.gateway < hlen)
359                         return 0;
360
361                 /* So... This guy found something strange INSIDE encapsulated
362                    packet. Well, he is fool, but what can we do ?
363                  */
364                 rel_type = ICMP_PARAMETERPROB;
365                 rel_info = skb->h.icmph->un.gateway - hlen;
366                 break;
367
368         case ICMP_DEST_UNREACH:
369                 switch (code) {
370                 case ICMP_SR_FAILED:
371                 case ICMP_PORT_UNREACH:
372                         /* Impossible event. */
373                         return 0;
374                 case ICMP_FRAG_NEEDED:
375                         /* And it is the only really necessary thing :-) */
376                         rel_info = ntohs(skb->h.icmph->un.frag.mtu);
377                         if (rel_info < hlen+68)
378                                 return 0;
379                         rel_info -= hlen;
380                         /* BSD 4.2 MORE DOES NOT EXIST IN NATURE. */
381                         if (rel_info > ntohs(eiph->tot_len))
382                                 return 0;
383                         break;
384                 default:
385                         /* All others are translated to HOST_UNREACH.
386                            rfc2003 contains "deep thoughts" about NET_UNREACH,
387                            I believe, it is just ether pollution. --ANK
388                          */
389                         rel_type = ICMP_DEST_UNREACH;
390                         rel_code = ICMP_HOST_UNREACH;
391                         break;
392                 }
393                 break;
394         case ICMP_TIME_EXCEEDED:
395                 if (code != ICMP_EXC_TTL)
396                         return 0;
397                 break;
398         }
399
400         /* Prepare fake skb to feed it to icmp_send */
401         skb2 = skb_clone(skb, GFP_ATOMIC);
402         if (skb2 == NULL)
403                 return 0;
404         dst_release(skb2->dst);
405         skb2->dst = NULL;
406         skb_pull(skb2, skb->data - (u8*)eiph);
407         skb2->nh.raw = skb2->data;
408
409         /* Try to guess incoming interface */
410         memset(&fl, 0, sizeof(fl));
411         fl.fl4_daddr = eiph->saddr;
412         fl.fl4_tos = RT_TOS(eiph->tos);
413         fl.proto = IPPROTO_IPIP;
414         if (ip_route_output_key(&rt, &key)) {
415                 kfree_skb(skb2);
416                 return 0;
417         }
418         skb2->dev = rt->u.dst.dev;
419
420         /* route "incoming" packet */
421         if (rt->rt_flags&RTCF_LOCAL) {
422                 ip_rt_put(rt);
423                 rt = NULL;
424                 fl.fl4_daddr = eiph->daddr;
425                 fl.fl4_src = eiph->saddr;
426                 fl.fl4_tos = eiph->tos;
427                 if (ip_route_output_key(&rt, &fl) ||
428                     rt->u.dst.dev->type != ARPHRD_TUNNEL) {
429                         ip_rt_put(rt);
430                         kfree_skb(skb2);
431                         return 0;
432                 }
433         } else {
434                 ip_rt_put(rt);
435                 if (ip_route_input(skb2, eiph->daddr, eiph->saddr, eiph->tos, skb2->dev) ||
436                     skb2->dst->dev->type != ARPHRD_TUNNEL) {
437                         kfree_skb(skb2);
438                         return 0;
439                 }
440         }
441
442         /* change mtu on this route */
443         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
444                 if (rel_info > dst_mtu(skb2->dst)) {
445                         kfree_skb(skb2);
446                         return 0;
447                 }
448                 skb2->dst->ops->update_pmtu(skb2->dst, rel_info);
449                 rel_info = htonl(rel_info);
450         } else if (type == ICMP_TIME_EXCEEDED) {
451                 struct ip_tunnel *t = netdev_priv(skb2->dev);
452                 if (t->parms.iph.ttl) {
453                         rel_type = ICMP_DEST_UNREACH;
454                         rel_code = ICMP_HOST_UNREACH;
455                 }
456         }
457
458         icmp_send(skb2, rel_type, rel_code, rel_info);
459         kfree_skb(skb2);
460         return 0;
461 #endif
462 }
463
464 static inline void ipip_ecn_decapsulate(struct iphdr *outer_iph, struct sk_buff *skb)
465 {
466         struct iphdr *inner_iph = skb->nh.iph;
467
468         if (INET_ECN_is_ce(outer_iph->tos))
469                 IP_ECN_set_ce(inner_iph);
470 }
471
472 static int ipip_rcv(struct sk_buff *skb)
473 {
474         struct iphdr *iph;
475         struct ip_tunnel *tunnel;
476
477         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
478                 goto out;
479
480         iph = skb->nh.iph;
481
482         read_lock(&ipip_lock);
483         if ((tunnel = ipip_tunnel_lookup(iph->saddr, iph->daddr)) != NULL) {
484                 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
485                         read_unlock(&ipip_lock);
486                         kfree_skb(skb);
487                         return 0;
488                 }
489
490                 secpath_reset(skb);
491
492                 skb->mac.raw = skb->nh.raw;
493                 skb->nh.raw = skb->data;
494                 memset(&(IPCB(skb)->opt), 0, sizeof(struct ip_options));
495                 skb->protocol = htons(ETH_P_IP);
496                 skb->pkt_type = PACKET_HOST;
497
498                 tunnel->stat.rx_packets++;
499                 tunnel->stat.rx_bytes += skb->len;
500                 skb->dev = tunnel->dev;
501                 dst_release(skb->dst);
502                 skb->dst = NULL;
503                 nf_reset(skb);
504                 ipip_ecn_decapsulate(iph, skb);
505                 netif_rx(skb);
506                 read_unlock(&ipip_lock);
507                 return 0;
508         }
509         read_unlock(&ipip_lock);
510
511 out:
512         return -1;
513 }
514
515 /*
516  *      This function assumes it is being called from dev_queue_xmit()
517  *      and that skb is filled properly by that function.
518  */
519
520 static int ipip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
521 {
522         struct ip_tunnel *tunnel = netdev_priv(dev);
523         struct net_device_stats *stats = &tunnel->stat;
524         struct iphdr  *tiph = &tunnel->parms.iph;
525         u8     tos = tunnel->parms.iph.tos;
526         u16    df = tiph->frag_off;
527         struct rtable *rt;                      /* Route to the other host */
528         struct net_device *tdev;                        /* Device to other host */
529         struct iphdr  *old_iph = skb->nh.iph;
530         struct iphdr  *iph;                     /* Our new IP header */
531         int    max_headroom;                    /* The extra header space needed */
532         u32    dst = tiph->daddr;
533         int    mtu;
534
535         if (tunnel->recursion++) {
536                 tunnel->stat.collisions++;
537                 goto tx_error;
538         }
539
540         if (skb->protocol != htons(ETH_P_IP))
541                 goto tx_error;
542
543         if (tos&1)
544                 tos = old_iph->tos;
545
546         if (!dst) {
547                 /* NBMA tunnel */
548                 if ((rt = (struct rtable*)skb->dst) == NULL) {
549                         tunnel->stat.tx_fifo_errors++;
550                         goto tx_error;
551                 }
552                 if ((dst = rt->rt_gateway) == 0)
553                         goto tx_error_icmp;
554         }
555
556         {
557                 struct flowi fl = { .oif = tunnel->parms.link,
558                                     .nl_u = { .ip4_u =
559                                               { .daddr = dst,
560                                                 .saddr = tiph->saddr,
561                                                 .tos = RT_TOS(tos) } },
562                                     .proto = IPPROTO_IPIP };
563                 if (ip_route_output_key(&rt, &fl)) {
564                         tunnel->stat.tx_carrier_errors++;
565                         goto tx_error_icmp;
566                 }
567         }
568         tdev = rt->u.dst.dev;
569
570         if (tdev == dev) {
571                 ip_rt_put(rt);
572                 tunnel->stat.collisions++;
573                 goto tx_error;
574         }
575
576         if (tiph->frag_off)
577                 mtu = dst_mtu(&rt->u.dst) - sizeof(struct iphdr);
578         else
579                 mtu = skb->dst ? dst_mtu(skb->dst) : dev->mtu;
580
581         if (mtu < 68) {
582                 tunnel->stat.collisions++;
583                 ip_rt_put(rt);
584                 goto tx_error;
585         }
586         if (skb->dst)
587                 skb->dst->ops->update_pmtu(skb->dst, mtu);
588
589         df |= (old_iph->frag_off&htons(IP_DF));
590
591         if ((old_iph->frag_off&htons(IP_DF)) && mtu < ntohs(old_iph->tot_len)) {
592                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
593                 ip_rt_put(rt);
594                 goto tx_error;
595         }
596
597         if (tunnel->err_count > 0) {
598                 if (jiffies - tunnel->err_time < IPTUNNEL_ERR_TIMEO) {
599                         tunnel->err_count--;
600                         dst_link_failure(skb);
601                 } else
602                         tunnel->err_count = 0;
603         }
604
605         /*
606          * Okay, now see if we can stuff it in the buffer as-is.
607          */
608         max_headroom = (LL_RESERVED_SPACE(tdev)+sizeof(struct iphdr));
609
610         if (skb_headroom(skb) < max_headroom || skb_cloned(skb) || skb_shared(skb)) {
611                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
612                 if (!new_skb) {
613                         ip_rt_put(rt);
614                         stats->tx_dropped++;
615                         dev_kfree_skb(skb);
616                         tunnel->recursion--;
617                         return 0;
618                 }
619                 if (skb->sk)
620                         skb_set_owner_w(new_skb, skb->sk);
621                 dev_kfree_skb(skb);
622                 skb = new_skb;
623                 old_iph = skb->nh.iph;
624         }
625
626         skb->h.raw = skb->nh.raw;
627         skb->nh.raw = skb_push(skb, sizeof(struct iphdr));
628         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
629         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
630                               IPSKB_REROUTED);
631         dst_release(skb->dst);
632         skb->dst = &rt->u.dst;
633
634         /*
635          *      Push down and install the IPIP header.
636          */
637
638         iph                     =       skb->nh.iph;
639         iph->version            =       4;
640         iph->ihl                =       sizeof(struct iphdr)>>2;
641         iph->frag_off           =       df;
642         iph->protocol           =       IPPROTO_IPIP;
643         iph->tos                =       INET_ECN_encapsulate(tos, old_iph->tos);
644         iph->daddr              =       rt->rt_dst;
645         iph->saddr              =       rt->rt_src;
646
647         if ((iph->ttl = tiph->ttl) == 0)
648                 iph->ttl        =       old_iph->ttl;
649
650         nf_reset(skb);
651
652         IPTUNNEL_XMIT();
653         tunnel->recursion--;
654         return 0;
655
656 tx_error_icmp:
657         dst_link_failure(skb);
658 tx_error:
659         stats->tx_errors++;
660         dev_kfree_skb(skb);
661         tunnel->recursion--;
662         return 0;
663 }
664
665 static int
666 ipip_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
667 {
668         int err = 0;
669         struct ip_tunnel_parm p;
670         struct ip_tunnel *t;
671
672         switch (cmd) {
673         case SIOCGETTUNNEL:
674                 t = NULL;
675                 if (dev == ipip_fb_tunnel_dev) {
676                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
677                                 err = -EFAULT;
678                                 break;
679                         }
680                         t = ipip_tunnel_locate(&p, 0);
681                 }
682                 if (t == NULL)
683                         t = netdev_priv(dev);
684                 memcpy(&p, &t->parms, sizeof(p));
685                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
686                         err = -EFAULT;
687                 break;
688
689         case SIOCADDTUNNEL:
690         case SIOCCHGTUNNEL:
691                 err = -EPERM;
692                 if (!capable(CAP_NET_ADMIN))
693                         goto done;
694
695                 err = -EFAULT;
696                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
697                         goto done;
698
699                 err = -EINVAL;
700                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPIP ||
701                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
702                         goto done;
703                 if (p.iph.ttl)
704                         p.iph.frag_off |= htons(IP_DF);
705
706                 t = ipip_tunnel_locate(&p, cmd == SIOCADDTUNNEL);
707
708                 if (dev != ipip_fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
709                         if (t != NULL) {
710                                 if (t->dev != dev) {
711                                         err = -EEXIST;
712                                         break;
713                                 }
714                         } else {
715                                 if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
716                                     (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
717                                         err = -EINVAL;
718                                         break;
719                                 }
720                                 t = netdev_priv(dev);
721                                 ipip_tunnel_unlink(t);
722                                 t->parms.iph.saddr = p.iph.saddr;
723                                 t->parms.iph.daddr = p.iph.daddr;
724                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
725                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
726                                 ipip_tunnel_link(t);
727                                 netdev_state_change(dev);
728                         }
729                 }
730
731                 if (t) {
732                         err = 0;
733                         if (cmd == SIOCCHGTUNNEL) {
734                                 t->parms.iph.ttl = p.iph.ttl;
735                                 t->parms.iph.tos = p.iph.tos;
736                                 t->parms.iph.frag_off = p.iph.frag_off;
737                         }
738                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
739                                 err = -EFAULT;
740                 } else
741                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
742                 break;
743
744         case SIOCDELTUNNEL:
745                 err = -EPERM;
746                 if (!capable(CAP_NET_ADMIN))
747                         goto done;
748
749                 if (dev == ipip_fb_tunnel_dev) {
750                         err = -EFAULT;
751                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
752                                 goto done;
753                         err = -ENOENT;
754                         if ((t = ipip_tunnel_locate(&p, 0)) == NULL)
755                                 goto done;
756                         err = -EPERM;
757                         if (t->dev == ipip_fb_tunnel_dev)
758                                 goto done;
759                         dev = t->dev;
760                 }
761                 err = unregister_netdevice(dev);
762                 break;
763
764         default:
765                 err = -EINVAL;
766         }
767
768 done:
769         return err;
770 }
771
772 static struct net_device_stats *ipip_tunnel_get_stats(struct net_device *dev)
773 {
774         return &(((struct ip_tunnel*)netdev_priv(dev))->stat);
775 }
776
777 static int ipip_tunnel_change_mtu(struct net_device *dev, int new_mtu)
778 {
779         if (new_mtu < 68 || new_mtu > 0xFFF8 - sizeof(struct iphdr))
780                 return -EINVAL;
781         dev->mtu = new_mtu;
782         return 0;
783 }
784
785 static void ipip_tunnel_setup(struct net_device *dev)
786 {
787         SET_MODULE_OWNER(dev);
788         dev->uninit             = ipip_tunnel_uninit;
789         dev->hard_start_xmit    = ipip_tunnel_xmit;
790         dev->get_stats          = ipip_tunnel_get_stats;
791         dev->do_ioctl           = ipip_tunnel_ioctl;
792         dev->change_mtu         = ipip_tunnel_change_mtu;
793         dev->destructor         = free_netdev;
794
795         dev->type               = ARPHRD_TUNNEL;
796         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr);
797         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr);
798         dev->flags              = IFF_NOARP;
799         dev->iflink             = 0;
800         dev->addr_len           = 4;
801 }
802
803 static int ipip_tunnel_init(struct net_device *dev)
804 {
805         struct net_device *tdev = NULL;
806         struct ip_tunnel *tunnel;
807         struct iphdr *iph;
808
809         tunnel = netdev_priv(dev);
810         iph = &tunnel->parms.iph;
811
812         tunnel->dev = dev;
813         strcpy(tunnel->parms.name, dev->name);
814
815         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
816         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
817
818         if (iph->daddr) {
819                 struct flowi fl = { .oif = tunnel->parms.link,
820                                     .nl_u = { .ip4_u =
821                                               { .daddr = iph->daddr,
822                                                 .saddr = iph->saddr,
823                                                 .tos = RT_TOS(iph->tos) } },
824                                     .proto = IPPROTO_IPIP };
825                 struct rtable *rt;
826                 if (!ip_route_output_key(&rt, &fl)) {
827                         tdev = rt->u.dst.dev;
828                         ip_rt_put(rt);
829                 }
830                 dev->flags |= IFF_POINTOPOINT;
831         }
832
833         if (!tdev && tunnel->parms.link)
834                 tdev = __dev_get_by_index(tunnel->parms.link);
835
836         if (tdev) {
837                 dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
838                 dev->mtu = tdev->mtu - sizeof(struct iphdr);
839         }
840         dev->iflink = tunnel->parms.link;
841
842         return 0;
843 }
844
845 static int __init ipip_fb_tunnel_init(struct net_device *dev)
846 {
847         struct ip_tunnel *tunnel = netdev_priv(dev);
848         struct iphdr *iph = &tunnel->parms.iph;
849
850         tunnel->dev = dev;
851         strcpy(tunnel->parms.name, dev->name);
852
853         iph->version            = 4;
854         iph->protocol           = IPPROTO_IPIP;
855         iph->ihl                = 5;
856
857         dev_hold(dev);
858         tunnels_wc[0]           = tunnel;
859         return 0;
860 }
861
862 static struct xfrm_tunnel ipip_handler = {
863         .handler        =       ipip_rcv,
864         .err_handler    =       ipip_err,
865         .priority       =       1,
866 };
867
868 static char banner[] __initdata =
869         KERN_INFO "IPv4 over IPv4 tunneling driver\n";
870
871 static int __init ipip_init(void)
872 {
873         int err;
874
875         printk(banner);
876
877         if (xfrm4_tunnel_register(&ipip_handler)) {
878                 printk(KERN_INFO "ipip init: can't register tunnel\n");
879                 return -EAGAIN;
880         }
881
882         ipip_fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel),
883                                            "tunl0",
884                                            ipip_tunnel_setup);
885         if (!ipip_fb_tunnel_dev) {
886                 err = -ENOMEM;
887                 goto err1;
888         }
889
890         ipip_fb_tunnel_dev->init = ipip_fb_tunnel_init;
891
892         if ((err = register_netdev(ipip_fb_tunnel_dev)))
893                 goto err2;
894  out:
895         return err;
896  err2:
897         free_netdev(ipip_fb_tunnel_dev);
898  err1:
899         xfrm4_tunnel_deregister(&ipip_handler);
900         goto out;
901 }
902
903 static void __exit ipip_destroy_tunnels(void)
904 {
905         int prio;
906
907         for (prio = 1; prio < 4; prio++) {
908                 int h;
909                 for (h = 0; h < HASH_SIZE; h++) {
910                         struct ip_tunnel *t;
911                         while ((t = tunnels[prio][h]) != NULL)
912                                 unregister_netdevice(t->dev);
913                 }
914         }
915 }
916
917 static void __exit ipip_fini(void)
918 {
919         if (xfrm4_tunnel_deregister(&ipip_handler))
920                 printk(KERN_INFO "ipip close: can't deregister tunnel\n");
921
922         rtnl_lock();
923         ipip_destroy_tunnels();
924         unregister_netdevice(ipip_fb_tunnel_dev);
925         rtnl_unlock();
926 }
927
928 module_init(ipip_init);
929 module_exit(ipip_fini);
930 MODULE_LICENSE("GPL");