netdev: convert pseudo-devices to netdev_tx_t
[linux-2.6.git] / 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@lxorguk.ukuu.org.uk) 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 void ipip_fb_tunnel_init(struct net_device *dev);
134 static void 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         nt->parms = *parms;
249
250         ipip_tunnel_init(dev);
251
252         if (register_netdevice(dev) < 0)
253                 goto failed_free;
254
255         dev_hold(dev);
256         ipip_tunnel_link(ipn, nt);
257         return nt;
258
259 failed_free:
260         free_netdev(dev);
261         return NULL;
262 }
263
264 static void ipip_tunnel_uninit(struct net_device *dev)
265 {
266         struct net *net = dev_net(dev);
267         struct ipip_net *ipn = net_generic(net, ipip_net_id);
268
269         if (dev == ipn->fb_tunnel_dev) {
270                 write_lock_bh(&ipip_lock);
271                 ipn->tunnels_wc[0] = NULL;
272                 write_unlock_bh(&ipip_lock);
273         } else
274                 ipip_tunnel_unlink(ipn, netdev_priv(dev));
275         dev_put(dev);
276 }
277
278 static int ipip_err(struct sk_buff *skb, u32 info)
279 {
280
281 /* All the routers (except for Linux) return only
282    8 bytes of packet payload. It means, that precise relaying of
283    ICMP in the real Internet is absolutely infeasible.
284  */
285         struct iphdr *iph = (struct iphdr *)skb->data;
286         const int type = icmp_hdr(skb)->type;
287         const int code = icmp_hdr(skb)->code;
288         struct ip_tunnel *t;
289         int err;
290
291         switch (type) {
292         default:
293         case ICMP_PARAMETERPROB:
294                 return 0;
295
296         case ICMP_DEST_UNREACH:
297                 switch (code) {
298                 case ICMP_SR_FAILED:
299                 case ICMP_PORT_UNREACH:
300                         /* Impossible event. */
301                         return 0;
302                 case ICMP_FRAG_NEEDED:
303                         /* Soft state for pmtu is maintained by IP core. */
304                         return 0;
305                 default:
306                         /* All others are translated to HOST_UNREACH.
307                            rfc2003 contains "deep thoughts" about NET_UNREACH,
308                            I believe they are just ether pollution. --ANK
309                          */
310                         break;
311                 }
312                 break;
313         case ICMP_TIME_EXCEEDED:
314                 if (code != ICMP_EXC_TTL)
315                         return 0;
316                 break;
317         }
318
319         err = -ENOENT;
320
321         read_lock(&ipip_lock);
322         t = ipip_tunnel_lookup(dev_net(skb->dev), iph->daddr, iph->saddr);
323         if (t == NULL || t->parms.iph.daddr == 0)
324                 goto out;
325
326         err = 0;
327         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
328                 goto out;
329
330         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
331                 t->err_count++;
332         else
333                 t->err_count = 1;
334         t->err_time = jiffies;
335 out:
336         read_unlock(&ipip_lock);
337         return err;
338 }
339
340 static inline void ipip_ecn_decapsulate(const struct iphdr *outer_iph,
341                                         struct sk_buff *skb)
342 {
343         struct iphdr *inner_iph = ip_hdr(skb);
344
345         if (INET_ECN_is_ce(outer_iph->tos))
346                 IP_ECN_set_ce(inner_iph);
347 }
348
349 static int ipip_rcv(struct sk_buff *skb)
350 {
351         struct ip_tunnel *tunnel;
352         const struct iphdr *iph = ip_hdr(skb);
353
354         read_lock(&ipip_lock);
355         if ((tunnel = ipip_tunnel_lookup(dev_net(skb->dev),
356                                         iph->saddr, iph->daddr)) != NULL) {
357                 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
358                         read_unlock(&ipip_lock);
359                         kfree_skb(skb);
360                         return 0;
361                 }
362
363                 secpath_reset(skb);
364
365                 skb->mac_header = skb->network_header;
366                 skb_reset_network_header(skb);
367                 skb->protocol = htons(ETH_P_IP);
368                 skb->pkt_type = PACKET_HOST;
369
370                 tunnel->dev->stats.rx_packets++;
371                 tunnel->dev->stats.rx_bytes += skb->len;
372                 skb->dev = tunnel->dev;
373                 skb_dst_drop(skb);
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 netdev_tx_t 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(skb)) == 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(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
450
451         if (mtu < 68) {
452                 stats->collisions++;
453                 ip_rt_put(rt);
454                 goto tx_error;
455         }
456         if (skb_dst(skb))
457                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), 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 (time_before(jiffies,
469                                 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
470                         tunnel->err_count--;
471                         dst_link_failure(skb);
472                 } else
473                         tunnel->err_count = 0;
474         }
475
476         /*
477          * Okay, now see if we can stuff it in the buffer as-is.
478          */
479         max_headroom = (LL_RESERVED_SPACE(tdev)+sizeof(struct iphdr));
480
481         if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
482             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
483                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
484                 if (!new_skb) {
485                         ip_rt_put(rt);
486                         stats->tx_dropped++;
487                         dev_kfree_skb(skb);
488                         tunnel->recursion--;
489                         return NETDEV_TX_OK;
490                 }
491                 if (skb->sk)
492                         skb_set_owner_w(new_skb, skb->sk);
493                 dev_kfree_skb(skb);
494                 skb = new_skb;
495                 old_iph = ip_hdr(skb);
496         }
497
498         skb->transport_header = skb->network_header;
499         skb_push(skb, sizeof(struct iphdr));
500         skb_reset_network_header(skb);
501         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
502         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
503                               IPSKB_REROUTED);
504         skb_dst_drop(skb);
505         skb_dst_set(skb, &rt->u.dst);
506
507         /*
508          *      Push down and install the IPIP header.
509          */
510
511         iph                     =       ip_hdr(skb);
512         iph->version            =       4;
513         iph->ihl                =       sizeof(struct iphdr)>>2;
514         iph->frag_off           =       df;
515         iph->protocol           =       IPPROTO_IPIP;
516         iph->tos                =       INET_ECN_encapsulate(tos, old_iph->tos);
517         iph->daddr              =       rt->rt_dst;
518         iph->saddr              =       rt->rt_src;
519
520         if ((iph->ttl = tiph->ttl) == 0)
521                 iph->ttl        =       old_iph->ttl;
522
523         nf_reset(skb);
524
525         IPTUNNEL_XMIT();
526         tunnel->recursion--;
527         return NETDEV_TX_OK;
528
529 tx_error_icmp:
530         dst_link_failure(skb);
531 tx_error:
532         stats->tx_errors++;
533         dev_kfree_skb(skb);
534         tunnel->recursion--;
535         return NETDEV_TX_OK;
536 }
537
538 static void ipip_tunnel_bind_dev(struct net_device *dev)
539 {
540         struct net_device *tdev = NULL;
541         struct ip_tunnel *tunnel;
542         struct iphdr *iph;
543
544         tunnel = netdev_priv(dev);
545         iph = &tunnel->parms.iph;
546
547         if (iph->daddr) {
548                 struct flowi fl = { .oif = tunnel->parms.link,
549                                     .nl_u = { .ip4_u =
550                                               { .daddr = iph->daddr,
551                                                 .saddr = iph->saddr,
552                                                 .tos = RT_TOS(iph->tos) } },
553                                     .proto = IPPROTO_IPIP };
554                 struct rtable *rt;
555                 if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
556                         tdev = rt->u.dst.dev;
557                         ip_rt_put(rt);
558                 }
559                 dev->flags |= IFF_POINTOPOINT;
560         }
561
562         if (!tdev && tunnel->parms.link)
563                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
564
565         if (tdev) {
566                 dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
567                 dev->mtu = tdev->mtu - sizeof(struct iphdr);
568         }
569         dev->iflink = tunnel->parms.link;
570 }
571
572 static int
573 ipip_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
574 {
575         int err = 0;
576         struct ip_tunnel_parm p;
577         struct ip_tunnel *t;
578         struct net *net = dev_net(dev);
579         struct ipip_net *ipn = net_generic(net, ipip_net_id);
580
581         switch (cmd) {
582         case SIOCGETTUNNEL:
583                 t = NULL;
584                 if (dev == ipn->fb_tunnel_dev) {
585                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
586                                 err = -EFAULT;
587                                 break;
588                         }
589                         t = ipip_tunnel_locate(net, &p, 0);
590                 }
591                 if (t == NULL)
592                         t = netdev_priv(dev);
593                 memcpy(&p, &t->parms, sizeof(p));
594                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
595                         err = -EFAULT;
596                 break;
597
598         case SIOCADDTUNNEL:
599         case SIOCCHGTUNNEL:
600                 err = -EPERM;
601                 if (!capable(CAP_NET_ADMIN))
602                         goto done;
603
604                 err = -EFAULT;
605                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
606                         goto done;
607
608                 err = -EINVAL;
609                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPIP ||
610                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
611                         goto done;
612                 if (p.iph.ttl)
613                         p.iph.frag_off |= htons(IP_DF);
614
615                 t = ipip_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
616
617                 if (dev != ipn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
618                         if (t != NULL) {
619                                 if (t->dev != dev) {
620                                         err = -EEXIST;
621                                         break;
622                                 }
623                         } else {
624                                 if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
625                                     (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
626                                         err = -EINVAL;
627                                         break;
628                                 }
629                                 t = netdev_priv(dev);
630                                 ipip_tunnel_unlink(ipn, t);
631                                 t->parms.iph.saddr = p.iph.saddr;
632                                 t->parms.iph.daddr = p.iph.daddr;
633                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
634                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
635                                 ipip_tunnel_link(ipn, t);
636                                 netdev_state_change(dev);
637                         }
638                 }
639
640                 if (t) {
641                         err = 0;
642                         if (cmd == SIOCCHGTUNNEL) {
643                                 t->parms.iph.ttl = p.iph.ttl;
644                                 t->parms.iph.tos = p.iph.tos;
645                                 t->parms.iph.frag_off = p.iph.frag_off;
646                                 if (t->parms.link != p.link) {
647                                         t->parms.link = p.link;
648                                         ipip_tunnel_bind_dev(dev);
649                                         netdev_state_change(dev);
650                                 }
651                         }
652                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
653                                 err = -EFAULT;
654                 } else
655                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
656                 break;
657
658         case SIOCDELTUNNEL:
659                 err = -EPERM;
660                 if (!capable(CAP_NET_ADMIN))
661                         goto done;
662
663                 if (dev == ipn->fb_tunnel_dev) {
664                         err = -EFAULT;
665                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
666                                 goto done;
667                         err = -ENOENT;
668                         if ((t = ipip_tunnel_locate(net, &p, 0)) == NULL)
669                                 goto done;
670                         err = -EPERM;
671                         if (t->dev == ipn->fb_tunnel_dev)
672                                 goto done;
673                         dev = t->dev;
674                 }
675                 unregister_netdevice(dev);
676                 err = 0;
677                 break;
678
679         default:
680                 err = -EINVAL;
681         }
682
683 done:
684         return err;
685 }
686
687 static int ipip_tunnel_change_mtu(struct net_device *dev, int new_mtu)
688 {
689         if (new_mtu < 68 || new_mtu > 0xFFF8 - sizeof(struct iphdr))
690                 return -EINVAL;
691         dev->mtu = new_mtu;
692         return 0;
693 }
694
695 static const struct net_device_ops ipip_netdev_ops = {
696         .ndo_uninit     = ipip_tunnel_uninit,
697         .ndo_start_xmit = ipip_tunnel_xmit,
698         .ndo_do_ioctl   = ipip_tunnel_ioctl,
699         .ndo_change_mtu = ipip_tunnel_change_mtu,
700
701 };
702
703 static void ipip_tunnel_setup(struct net_device *dev)
704 {
705         dev->netdev_ops         = &ipip_netdev_ops;
706         dev->destructor         = free_netdev;
707
708         dev->type               = ARPHRD_TUNNEL;
709         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr);
710         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr);
711         dev->flags              = IFF_NOARP;
712         dev->iflink             = 0;
713         dev->addr_len           = 4;
714         dev->features           |= NETIF_F_NETNS_LOCAL;
715         dev->priv_flags         &= ~IFF_XMIT_DST_RELEASE;
716 }
717
718 static void ipip_tunnel_init(struct net_device *dev)
719 {
720         struct ip_tunnel *tunnel = netdev_priv(dev);
721
722         tunnel->dev = dev;
723         strcpy(tunnel->parms.name, dev->name);
724
725         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
726         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
727
728         ipip_tunnel_bind_dev(dev);
729 }
730
731 static void ipip_fb_tunnel_init(struct net_device *dev)
732 {
733         struct ip_tunnel *tunnel = netdev_priv(dev);
734         struct iphdr *iph = &tunnel->parms.iph;
735         struct ipip_net *ipn = net_generic(dev_net(dev), ipip_net_id);
736
737         tunnel->dev = dev;
738         strcpy(tunnel->parms.name, dev->name);
739
740         iph->version            = 4;
741         iph->protocol           = IPPROTO_IPIP;
742         iph->ihl                = 5;
743
744         dev_hold(dev);
745         ipn->tunnels_wc[0]      = tunnel;
746 }
747
748 static struct xfrm_tunnel ipip_handler = {
749         .handler        =       ipip_rcv,
750         .err_handler    =       ipip_err,
751         .priority       =       1,
752 };
753
754 static const char banner[] __initconst =
755         KERN_INFO "IPv4 over IPv4 tunneling driver\n";
756
757 static void ipip_destroy_tunnels(struct ipip_net *ipn)
758 {
759         int prio;
760
761         for (prio = 1; prio < 4; prio++) {
762                 int h;
763                 for (h = 0; h < HASH_SIZE; h++) {
764                         struct ip_tunnel *t;
765                         while ((t = ipn->tunnels[prio][h]) != NULL)
766                                 unregister_netdevice(t->dev);
767                 }
768         }
769 }
770
771 static int ipip_init_net(struct net *net)
772 {
773         int err;
774         struct ipip_net *ipn;
775
776         err = -ENOMEM;
777         ipn = kzalloc(sizeof(struct ipip_net), GFP_KERNEL);
778         if (ipn == NULL)
779                 goto err_alloc;
780
781         err = net_assign_generic(net, ipip_net_id, ipn);
782         if (err < 0)
783                 goto err_assign;
784
785         ipn->tunnels[0] = ipn->tunnels_wc;
786         ipn->tunnels[1] = ipn->tunnels_l;
787         ipn->tunnels[2] = ipn->tunnels_r;
788         ipn->tunnels[3] = ipn->tunnels_r_l;
789
790         ipn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel),
791                                            "tunl0",
792                                            ipip_tunnel_setup);
793         if (!ipn->fb_tunnel_dev) {
794                 err = -ENOMEM;
795                 goto err_alloc_dev;
796         }
797         dev_net_set(ipn->fb_tunnel_dev, net);
798
799         ipip_fb_tunnel_init(ipn->fb_tunnel_dev);
800
801         if ((err = register_netdev(ipn->fb_tunnel_dev)))
802                 goto err_reg_dev;
803
804         return 0;
805
806 err_reg_dev:
807         free_netdev(ipn->fb_tunnel_dev);
808 err_alloc_dev:
809         /* nothing */
810 err_assign:
811         kfree(ipn);
812 err_alloc:
813         return err;
814 }
815
816 static void ipip_exit_net(struct net *net)
817 {
818         struct ipip_net *ipn;
819
820         ipn = net_generic(net, ipip_net_id);
821         rtnl_lock();
822         ipip_destroy_tunnels(ipn);
823         unregister_netdevice(ipn->fb_tunnel_dev);
824         rtnl_unlock();
825         kfree(ipn);
826 }
827
828 static struct pernet_operations ipip_net_ops = {
829         .init = ipip_init_net,
830         .exit = ipip_exit_net,
831 };
832
833 static int __init ipip_init(void)
834 {
835         int err;
836
837         printk(banner);
838
839         if (xfrm4_tunnel_register(&ipip_handler, AF_INET)) {
840                 printk(KERN_INFO "ipip init: can't register tunnel\n");
841                 return -EAGAIN;
842         }
843
844         err = register_pernet_gen_device(&ipip_net_id, &ipip_net_ops);
845         if (err)
846                 xfrm4_tunnel_deregister(&ipip_handler, AF_INET);
847
848         return err;
849 }
850
851 static void __exit ipip_fini(void)
852 {
853         if (xfrm4_tunnel_deregister(&ipip_handler, AF_INET))
854                 printk(KERN_INFO "ipip close: can't deregister tunnel\n");
855
856         unregister_pernet_gen_device(ipip_net_id, &ipip_net_ops);
857 }
858
859 module_init(ipip_init);
860 module_exit(ipip_fini);
861 MODULE_LICENSE("GPL");