raw: Raw socket leak.
[linux-2.6.git] / net / ipv6 / raw.c
1 /*
2  *      RAW sockets for IPv6
3  *      Linux INET6 implementation
4  *
5  *      Authors:
6  *      Pedro Roque             <roque@di.fc.ul.pt>
7  *
8  *      Adapted from linux/net/ipv4/raw.c
9  *
10  *      $Id: raw.c,v 1.51 2002/02/01 22:01:04 davem Exp $
11  *
12  *      Fixes:
13  *      Hideaki YOSHIFUJI       :       sin6_scope_id support
14  *      YOSHIFUJI,H.@USAGI      :       raw checksum (RFC2292(bis) compliance)
15  *      Kazunori MIYAZAWA @USAGI:       change process style to use ip6_append_data
16  *
17  *      This program is free software; you can redistribute it and/or
18  *      modify it under the terms of the GNU General Public License
19  *      as published by the Free Software Foundation; either version
20  *      2 of the License, or (at your option) any later version.
21  */
22
23 #include <linux/errno.h>
24 #include <linux/types.h>
25 #include <linux/socket.h>
26 #include <linux/sockios.h>
27 #include <linux/net.h>
28 #include <linux/in6.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_arp.h>
31 #include <linux/icmpv6.h>
32 #include <linux/netfilter.h>
33 #include <linux/netfilter_ipv6.h>
34 #include <linux/skbuff.h>
35 #include <asm/uaccess.h>
36 #include <asm/ioctls.h>
37
38 #include <net/net_namespace.h>
39 #include <net/ip.h>
40 #include <net/sock.h>
41 #include <net/snmp.h>
42
43 #include <net/ipv6.h>
44 #include <net/ndisc.h>
45 #include <net/protocol.h>
46 #include <net/ip6_route.h>
47 #include <net/ip6_checksum.h>
48 #include <net/addrconf.h>
49 #include <net/transp_v6.h>
50 #include <net/udp.h>
51 #include <net/inet_common.h>
52 #include <net/tcp_states.h>
53 #if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
54 #include <net/mip6.h>
55 #endif
56 #include <linux/mroute6.h>
57
58 #include <net/raw.h>
59 #include <net/rawv6.h>
60 #include <net/xfrm.h>
61
62 #include <linux/proc_fs.h>
63 #include <linux/seq_file.h>
64
65 static struct raw_hashinfo raw_v6_hashinfo = {
66         .lock = __RW_LOCK_UNLOCKED(raw_v6_hashinfo.lock),
67 };
68
69 static struct sock *__raw_v6_lookup(struct net *net, struct sock *sk,
70                 unsigned short num, struct in6_addr *loc_addr,
71                 struct in6_addr *rmt_addr, int dif)
72 {
73         struct hlist_node *node;
74         int is_multicast = ipv6_addr_is_multicast(loc_addr);
75
76         sk_for_each_from(sk, node)
77                 if (inet_sk(sk)->num == num) {
78                         struct ipv6_pinfo *np = inet6_sk(sk);
79
80                         if (!net_eq(sock_net(sk), net))
81                                 continue;
82
83                         if (!ipv6_addr_any(&np->daddr) &&
84                             !ipv6_addr_equal(&np->daddr, rmt_addr))
85                                 continue;
86
87                         if (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif)
88                                 continue;
89
90                         if (!ipv6_addr_any(&np->rcv_saddr)) {
91                                 if (ipv6_addr_equal(&np->rcv_saddr, loc_addr))
92                                         goto found;
93                                 if (is_multicast &&
94                                     inet6_mc_check(sk, loc_addr, rmt_addr))
95                                         goto found;
96                                 continue;
97                         }
98                         goto found;
99                 }
100         sk = NULL;
101 found:
102         return sk;
103 }
104
105 /*
106  *      0 - deliver
107  *      1 - block
108  */
109 static __inline__ int icmpv6_filter(struct sock *sk, struct sk_buff *skb)
110 {
111         struct icmp6hdr *icmph;
112         struct raw6_sock *rp = raw6_sk(sk);
113
114         if (pskb_may_pull(skb, sizeof(struct icmp6hdr))) {
115                 __u32 *data = &rp->filter.data[0];
116                 int bit_nr;
117
118                 icmph = (struct icmp6hdr *) skb->data;
119                 bit_nr = icmph->icmp6_type;
120
121                 return (data[bit_nr >> 5] & (1 << (bit_nr & 31))) != 0;
122         }
123         return 0;
124 }
125
126 #if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
127 static int (*mh_filter)(struct sock *sock, struct sk_buff *skb);
128
129 int rawv6_mh_filter_register(int (*filter)(struct sock *sock,
130                                            struct sk_buff *skb))
131 {
132         rcu_assign_pointer(mh_filter, filter);
133         return 0;
134 }
135 EXPORT_SYMBOL(rawv6_mh_filter_register);
136
137 int rawv6_mh_filter_unregister(int (*filter)(struct sock *sock,
138                                              struct sk_buff *skb))
139 {
140         rcu_assign_pointer(mh_filter, NULL);
141         synchronize_rcu();
142         return 0;
143 }
144 EXPORT_SYMBOL(rawv6_mh_filter_unregister);
145
146 #endif
147
148 /*
149  *      demultiplex raw sockets.
150  *      (should consider queueing the skb in the sock receive_queue
151  *      without calling rawv6.c)
152  *
153  *      Caller owns SKB so we must make clones.
154  */
155 static int ipv6_raw_deliver(struct sk_buff *skb, int nexthdr)
156 {
157         struct in6_addr *saddr;
158         struct in6_addr *daddr;
159         struct sock *sk;
160         int delivered = 0;
161         __u8 hash;
162         struct net *net;
163
164         saddr = &ipv6_hdr(skb)->saddr;
165         daddr = saddr + 1;
166
167         hash = nexthdr & (MAX_INET_PROTOS - 1);
168
169         read_lock(&raw_v6_hashinfo.lock);
170         sk = sk_head(&raw_v6_hashinfo.ht[hash]);
171
172         if (sk == NULL)
173                 goto out;
174
175         net = dev_net(skb->dev);
176         sk = __raw_v6_lookup(net, sk, nexthdr, daddr, saddr, IP6CB(skb)->iif);
177
178         while (sk) {
179                 int filtered;
180
181                 delivered = 1;
182                 switch (nexthdr) {
183                 case IPPROTO_ICMPV6:
184                         filtered = icmpv6_filter(sk, skb);
185                         break;
186
187 #if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
188                 case IPPROTO_MH:
189                 {
190                         /* XXX: To validate MH only once for each packet,
191                          * this is placed here. It should be after checking
192                          * xfrm policy, however it doesn't. The checking xfrm
193                          * policy is placed in rawv6_rcv() because it is
194                          * required for each socket.
195                          */
196                         int (*filter)(struct sock *sock, struct sk_buff *skb);
197
198                         filter = rcu_dereference(mh_filter);
199                         filtered = filter ? filter(sk, skb) : 0;
200                         break;
201                 }
202 #endif
203                 default:
204                         filtered = 0;
205                         break;
206                 }
207
208                 if (filtered < 0)
209                         break;
210                 if (filtered == 0) {
211                         struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
212
213                         /* Not releasing hash table! */
214                         if (clone) {
215                                 nf_reset(clone);
216                                 rawv6_rcv(sk, clone);
217                         }
218                 }
219                 sk = __raw_v6_lookup(net, sk_next(sk), nexthdr, daddr, saddr,
220                                      IP6CB(skb)->iif);
221         }
222 out:
223         read_unlock(&raw_v6_hashinfo.lock);
224         return delivered;
225 }
226
227 int raw6_local_deliver(struct sk_buff *skb, int nexthdr)
228 {
229         struct sock *raw_sk;
230
231         raw_sk = sk_head(&raw_v6_hashinfo.ht[nexthdr & (MAX_INET_PROTOS - 1)]);
232         if (raw_sk && !ipv6_raw_deliver(skb, nexthdr))
233                 raw_sk = NULL;
234
235         return raw_sk != NULL;
236 }
237
238 /* This cleans up af_inet6 a bit. -DaveM */
239 static int rawv6_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
240 {
241         struct inet_sock *inet = inet_sk(sk);
242         struct ipv6_pinfo *np = inet6_sk(sk);
243         struct sockaddr_in6 *addr = (struct sockaddr_in6 *) uaddr;
244         __be32 v4addr = 0;
245         int addr_type;
246         int err;
247
248         if (addr_len < SIN6_LEN_RFC2133)
249                 return -EINVAL;
250         addr_type = ipv6_addr_type(&addr->sin6_addr);
251
252         /* Raw sockets are IPv6 only */
253         if (addr_type == IPV6_ADDR_MAPPED)
254                 return(-EADDRNOTAVAIL);
255
256         lock_sock(sk);
257
258         err = -EINVAL;
259         if (sk->sk_state != TCP_CLOSE)
260                 goto out;
261
262         /* Check if the address belongs to the host. */
263         if (addr_type != IPV6_ADDR_ANY) {
264                 struct net_device *dev = NULL;
265
266                 if (addr_type & IPV6_ADDR_LINKLOCAL) {
267                         if (addr_len >= sizeof(struct sockaddr_in6) &&
268                             addr->sin6_scope_id) {
269                                 /* Override any existing binding, if another
270                                  * one is supplied by user.
271                                  */
272                                 sk->sk_bound_dev_if = addr->sin6_scope_id;
273                         }
274
275                         /* Binding to link-local address requires an interface */
276                         if (!sk->sk_bound_dev_if)
277                                 goto out;
278
279                         dev = dev_get_by_index(sock_net(sk), sk->sk_bound_dev_if);
280                         if (!dev) {
281                                 err = -ENODEV;
282                                 goto out;
283                         }
284                 }
285
286                 /* ipv4 addr of the socket is invalid.  Only the
287                  * unspecified and mapped address have a v4 equivalent.
288                  */
289                 v4addr = LOOPBACK4_IPV6;
290                 if (!(addr_type & IPV6_ADDR_MULTICAST)) {
291                         err = -EADDRNOTAVAIL;
292                         if (!ipv6_chk_addr(sock_net(sk), &addr->sin6_addr,
293                                            dev, 0)) {
294                                 if (dev)
295                                         dev_put(dev);
296                                 goto out;
297                         }
298                 }
299                 if (dev)
300                         dev_put(dev);
301         }
302
303         inet->rcv_saddr = inet->saddr = v4addr;
304         ipv6_addr_copy(&np->rcv_saddr, &addr->sin6_addr);
305         if (!(addr_type & IPV6_ADDR_MULTICAST))
306                 ipv6_addr_copy(&np->saddr, &addr->sin6_addr);
307         err = 0;
308 out:
309         release_sock(sk);
310         return err;
311 }
312
313 static void rawv6_err(struct sock *sk, struct sk_buff *skb,
314                struct inet6_skb_parm *opt,
315                int type, int code, int offset, __be32 info)
316 {
317         struct inet_sock *inet = inet_sk(sk);
318         struct ipv6_pinfo *np = inet6_sk(sk);
319         int err;
320         int harderr;
321
322         /* Report error on raw socket, if:
323            1. User requested recverr.
324            2. Socket is connected (otherwise the error indication
325               is useless without recverr and error is hard.
326          */
327         if (!np->recverr && sk->sk_state != TCP_ESTABLISHED)
328                 return;
329
330         harderr = icmpv6_err_convert(type, code, &err);
331         if (type == ICMPV6_PKT_TOOBIG)
332                 harderr = (np->pmtudisc == IPV6_PMTUDISC_DO);
333
334         if (np->recverr) {
335                 u8 *payload = skb->data;
336                 if (!inet->hdrincl)
337                         payload += offset;
338                 ipv6_icmp_error(sk, skb, err, 0, ntohl(info), payload);
339         }
340
341         if (np->recverr || harderr) {
342                 sk->sk_err = err;
343                 sk->sk_error_report(sk);
344         }
345 }
346
347 void raw6_icmp_error(struct sk_buff *skb, int nexthdr,
348                 int type, int code, int inner_offset, __be32 info)
349 {
350         struct sock *sk;
351         int hash;
352         struct in6_addr *saddr, *daddr;
353         struct net *net;
354
355         hash = nexthdr & (RAW_HTABLE_SIZE - 1);
356
357         read_lock(&raw_v6_hashinfo.lock);
358         sk = sk_head(&raw_v6_hashinfo.ht[hash]);
359         if (sk != NULL) {
360                 /* Note: ipv6_hdr(skb) != skb->data */
361                 struct ipv6hdr *ip6h = (struct ipv6hdr *)skb->data;
362                 saddr = &ip6h->saddr;
363                 daddr = &ip6h->daddr;
364                 net = dev_net(skb->dev);
365
366                 while ((sk = __raw_v6_lookup(net, sk, nexthdr, saddr, daddr,
367                                                 IP6CB(skb)->iif))) {
368                         rawv6_err(sk, skb, NULL, type, code,
369                                         inner_offset, info);
370                         sk = sk_next(sk);
371                 }
372         }
373         read_unlock(&raw_v6_hashinfo.lock);
374 }
375
376 static inline int rawv6_rcv_skb(struct sock * sk, struct sk_buff * skb)
377 {
378         if ((raw6_sk(sk)->checksum || sk->sk_filter) &&
379             skb_checksum_complete(skb)) {
380                 atomic_inc(&sk->sk_drops);
381                 kfree_skb(skb);
382                 return 0;
383         }
384
385         /* Charge it to the socket. */
386         if (sock_queue_rcv_skb(sk,skb)<0) {
387                 atomic_inc(&sk->sk_drops);
388                 kfree_skb(skb);
389                 return 0;
390         }
391
392         return 0;
393 }
394
395 /*
396  *      This is next to useless...
397  *      if we demultiplex in network layer we don't need the extra call
398  *      just to queue the skb...
399  *      maybe we could have the network decide upon a hint if it
400  *      should call raw_rcv for demultiplexing
401  */
402 int rawv6_rcv(struct sock *sk, struct sk_buff *skb)
403 {
404         struct inet_sock *inet = inet_sk(sk);
405         struct raw6_sock *rp = raw6_sk(sk);
406
407         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
408                 atomic_inc(&sk->sk_drops);
409                 kfree_skb(skb);
410                 return NET_RX_DROP;
411         }
412
413         if (!rp->checksum)
414                 skb->ip_summed = CHECKSUM_UNNECESSARY;
415
416         if (skb->ip_summed == CHECKSUM_COMPLETE) {
417                 skb_postpull_rcsum(skb, skb_network_header(skb),
418                                    skb_network_header_len(skb));
419                 if (!csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
420                                      &ipv6_hdr(skb)->daddr,
421                                      skb->len, inet->num, skb->csum))
422                         skb->ip_summed = CHECKSUM_UNNECESSARY;
423         }
424         if (!skb_csum_unnecessary(skb))
425                 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
426                                                          &ipv6_hdr(skb)->daddr,
427                                                          skb->len,
428                                                          inet->num, 0));
429
430         if (inet->hdrincl) {
431                 if (skb_checksum_complete(skb)) {
432                         atomic_inc(&sk->sk_drops);
433                         kfree_skb(skb);
434                         return 0;
435                 }
436         }
437
438         rawv6_rcv_skb(sk, skb);
439         return 0;
440 }
441
442
443 /*
444  *      This should be easy, if there is something there
445  *      we return it, otherwise we block.
446  */
447
448 static int rawv6_recvmsg(struct kiocb *iocb, struct sock *sk,
449                   struct msghdr *msg, size_t len,
450                   int noblock, int flags, int *addr_len)
451 {
452         struct ipv6_pinfo *np = inet6_sk(sk);
453         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)msg->msg_name;
454         struct sk_buff *skb;
455         size_t copied;
456         int err;
457
458         if (flags & MSG_OOB)
459                 return -EOPNOTSUPP;
460
461         if (addr_len)
462                 *addr_len=sizeof(*sin6);
463
464         if (flags & MSG_ERRQUEUE)
465                 return ipv6_recv_error(sk, msg, len);
466
467         skb = skb_recv_datagram(sk, flags, noblock, &err);
468         if (!skb)
469                 goto out;
470
471         copied = skb->len;
472         if (copied > len) {
473                 copied = len;
474                 msg->msg_flags |= MSG_TRUNC;
475         }
476
477         if (skb_csum_unnecessary(skb)) {
478                 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
479         } else if (msg->msg_flags&MSG_TRUNC) {
480                 if (__skb_checksum_complete(skb))
481                         goto csum_copy_err;
482                 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
483         } else {
484                 err = skb_copy_and_csum_datagram_iovec(skb, 0, msg->msg_iov);
485                 if (err == -EINVAL)
486                         goto csum_copy_err;
487         }
488         if (err)
489                 goto out_free;
490
491         /* Copy the address. */
492         if (sin6) {
493                 sin6->sin6_family = AF_INET6;
494                 sin6->sin6_port = 0;
495                 ipv6_addr_copy(&sin6->sin6_addr, &ipv6_hdr(skb)->saddr);
496                 sin6->sin6_flowinfo = 0;
497                 sin6->sin6_scope_id = 0;
498                 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
499                         sin6->sin6_scope_id = IP6CB(skb)->iif;
500         }
501
502         sock_recv_timestamp(msg, sk, skb);
503
504         if (np->rxopt.all)
505                 datagram_recv_ctl(sk, msg, skb);
506
507         err = copied;
508         if (flags & MSG_TRUNC)
509                 err = skb->len;
510
511 out_free:
512         skb_free_datagram(sk, skb);
513 out:
514         return err;
515
516 csum_copy_err:
517         skb_kill_datagram(sk, skb, flags);
518
519         /* Error for blocking case is chosen to masquerade
520            as some normal condition.
521          */
522         err = (flags&MSG_DONTWAIT) ? -EAGAIN : -EHOSTUNREACH;
523         atomic_inc(&sk->sk_drops);
524         goto out;
525 }
526
527 static int rawv6_push_pending_frames(struct sock *sk, struct flowi *fl,
528                                      struct raw6_sock *rp)
529 {
530         struct sk_buff *skb;
531         int err = 0;
532         int offset;
533         int len;
534         int total_len;
535         __wsum tmp_csum;
536         __sum16 csum;
537
538         if (!rp->checksum)
539                 goto send;
540
541         if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
542                 goto out;
543
544         offset = rp->offset;
545         total_len = inet_sk(sk)->cork.length - (skb_network_header(skb) -
546                                                 skb->data);
547         if (offset >= total_len - 1) {
548                 err = -EINVAL;
549                 ip6_flush_pending_frames(sk);
550                 goto out;
551         }
552
553         /* should be check HW csum miyazawa */
554         if (skb_queue_len(&sk->sk_write_queue) == 1) {
555                 /*
556                  * Only one fragment on the socket.
557                  */
558                 tmp_csum = skb->csum;
559         } else {
560                 struct sk_buff *csum_skb = NULL;
561                 tmp_csum = 0;
562
563                 skb_queue_walk(&sk->sk_write_queue, skb) {
564                         tmp_csum = csum_add(tmp_csum, skb->csum);
565
566                         if (csum_skb)
567                                 continue;
568
569                         len = skb->len - skb_transport_offset(skb);
570                         if (offset >= len) {
571                                 offset -= len;
572                                 continue;
573                         }
574
575                         csum_skb = skb;
576                 }
577
578                 skb = csum_skb;
579         }
580
581         offset += skb_transport_offset(skb);
582         if (skb_copy_bits(skb, offset, &csum, 2))
583                 BUG();
584
585         /* in case cksum was not initialized */
586         if (unlikely(csum))
587                 tmp_csum = csum_sub(tmp_csum, csum_unfold(csum));
588
589         csum = csum_ipv6_magic(&fl->fl6_src,
590                                    &fl->fl6_dst,
591                                    total_len, fl->proto, tmp_csum);
592
593         if (csum == 0 && fl->proto == IPPROTO_UDP)
594                 csum = CSUM_MANGLED_0;
595
596         if (skb_store_bits(skb, offset, &csum, 2))
597                 BUG();
598
599 send:
600         err = ip6_push_pending_frames(sk);
601 out:
602         return err;
603 }
604
605 static int rawv6_send_hdrinc(struct sock *sk, void *from, int length,
606                         struct flowi *fl, struct rt6_info *rt,
607                         unsigned int flags)
608 {
609         struct ipv6_pinfo *np = inet6_sk(sk);
610         struct ipv6hdr *iph;
611         struct sk_buff *skb;
612         int err;
613
614         if (length > rt->u.dst.dev->mtu) {
615                 ipv6_local_error(sk, EMSGSIZE, fl, rt->u.dst.dev->mtu);
616                 return -EMSGSIZE;
617         }
618         if (flags&MSG_PROBE)
619                 goto out;
620
621         skb = sock_alloc_send_skb(sk,
622                                   length + LL_ALLOCATED_SPACE(rt->u.dst.dev) + 15,
623                                   flags & MSG_DONTWAIT, &err);
624         if (skb == NULL)
625                 goto error;
626         skb_reserve(skb, LL_RESERVED_SPACE(rt->u.dst.dev));
627
628         skb->priority = sk->sk_priority;
629         skb->mark = sk->sk_mark;
630         skb->dst = dst_clone(&rt->u.dst);
631
632         skb_put(skb, length);
633         skb_reset_network_header(skb);
634         iph = ipv6_hdr(skb);
635
636         skb->ip_summed = CHECKSUM_NONE;
637
638         skb->transport_header = skb->network_header;
639         err = memcpy_fromiovecend((void *)iph, from, 0, length);
640         if (err)
641                 goto error_fault;
642
643         IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
644         err = NF_HOOK(PF_INET6, NF_INET_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
645                       dst_output);
646         if (err > 0)
647                 err = np->recverr ? net_xmit_errno(err) : 0;
648         if (err)
649                 goto error;
650 out:
651         return 0;
652
653 error_fault:
654         err = -EFAULT;
655         kfree_skb(skb);
656 error:
657         IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
658         return err;
659 }
660
661 static int rawv6_probe_proto_opt(struct flowi *fl, struct msghdr *msg)
662 {
663         struct iovec *iov;
664         u8 __user *type = NULL;
665         u8 __user *code = NULL;
666         u8 len = 0;
667         int probed = 0;
668         int i;
669
670         if (!msg->msg_iov)
671                 return 0;
672
673         for (i = 0; i < msg->msg_iovlen; i++) {
674                 iov = &msg->msg_iov[i];
675                 if (!iov)
676                         continue;
677
678                 switch (fl->proto) {
679                 case IPPROTO_ICMPV6:
680                         /* check if one-byte field is readable or not. */
681                         if (iov->iov_base && iov->iov_len < 1)
682                                 break;
683
684                         if (!type) {
685                                 type = iov->iov_base;
686                                 /* check if code field is readable or not. */
687                                 if (iov->iov_len > 1)
688                                         code = type + 1;
689                         } else if (!code)
690                                 code = iov->iov_base;
691
692                         if (type && code) {
693                                 if (get_user(fl->fl_icmp_type, type) ||
694                                     get_user(fl->fl_icmp_code, code))
695                                         return -EFAULT;
696                                 probed = 1;
697                         }
698                         break;
699                 case IPPROTO_MH:
700                         if (iov->iov_base && iov->iov_len < 1)
701                                 break;
702                         /* check if type field is readable or not. */
703                         if (iov->iov_len > 2 - len) {
704                                 u8 __user *p = iov->iov_base;
705                                 if (get_user(fl->fl_mh_type, &p[2 - len]))
706                                         return -EFAULT;
707                                 probed = 1;
708                         } else
709                                 len += iov->iov_len;
710
711                         break;
712                 default:
713                         probed = 1;
714                         break;
715                 }
716                 if (probed)
717                         break;
718         }
719         return 0;
720 }
721
722 static int rawv6_sendmsg(struct kiocb *iocb, struct sock *sk,
723                    struct msghdr *msg, size_t len)
724 {
725         struct ipv6_txoptions opt_space;
726         struct sockaddr_in6 * sin6 = (struct sockaddr_in6 *) msg->msg_name;
727         struct in6_addr *daddr, *final_p = NULL, final;
728         struct inet_sock *inet = inet_sk(sk);
729         struct ipv6_pinfo *np = inet6_sk(sk);
730         struct raw6_sock *rp = raw6_sk(sk);
731         struct ipv6_txoptions *opt = NULL;
732         struct ip6_flowlabel *flowlabel = NULL;
733         struct dst_entry *dst = NULL;
734         struct flowi fl;
735         int addr_len = msg->msg_namelen;
736         int hlimit = -1;
737         int tclass = -1;
738         u16 proto;
739         int err;
740
741         /* Rough check on arithmetic overflow,
742            better check is made in ip6_append_data().
743          */
744         if (len > INT_MAX)
745                 return -EMSGSIZE;
746
747         /* Mirror BSD error message compatibility */
748         if (msg->msg_flags & MSG_OOB)
749                 return -EOPNOTSUPP;
750
751         /*
752          *      Get and verify the address.
753          */
754         memset(&fl, 0, sizeof(fl));
755
756         fl.mark = sk->sk_mark;
757
758         if (sin6) {
759                 if (addr_len < SIN6_LEN_RFC2133)
760                         return -EINVAL;
761
762                 if (sin6->sin6_family && sin6->sin6_family != AF_INET6)
763                         return(-EAFNOSUPPORT);
764
765                 /* port is the proto value [0..255] carried in nexthdr */
766                 proto = ntohs(sin6->sin6_port);
767
768                 if (!proto)
769                         proto = inet->num;
770                 else if (proto != inet->num)
771                         return(-EINVAL);
772
773                 if (proto > 255)
774                         return(-EINVAL);
775
776                 daddr = &sin6->sin6_addr;
777                 if (np->sndflow) {
778                         fl.fl6_flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
779                         if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
780                                 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
781                                 if (flowlabel == NULL)
782                                         return -EINVAL;
783                                 daddr = &flowlabel->dst;
784                         }
785                 }
786
787                 /*
788                  * Otherwise it will be difficult to maintain
789                  * sk->sk_dst_cache.
790                  */
791                 if (sk->sk_state == TCP_ESTABLISHED &&
792                     ipv6_addr_equal(daddr, &np->daddr))
793                         daddr = &np->daddr;
794
795                 if (addr_len >= sizeof(struct sockaddr_in6) &&
796                     sin6->sin6_scope_id &&
797                     ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
798                         fl.oif = sin6->sin6_scope_id;
799         } else {
800                 if (sk->sk_state != TCP_ESTABLISHED)
801                         return -EDESTADDRREQ;
802
803                 proto = inet->num;
804                 daddr = &np->daddr;
805                 fl.fl6_flowlabel = np->flow_label;
806         }
807
808         if (fl.oif == 0)
809                 fl.oif = sk->sk_bound_dev_if;
810
811         if (msg->msg_controllen) {
812                 opt = &opt_space;
813                 memset(opt, 0, sizeof(struct ipv6_txoptions));
814                 opt->tot_len = sizeof(struct ipv6_txoptions);
815
816                 err = datagram_send_ctl(sock_net(sk), msg, &fl, opt, &hlimit, &tclass);
817                 if (err < 0) {
818                         fl6_sock_release(flowlabel);
819                         return err;
820                 }
821                 if ((fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
822                         flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
823                         if (flowlabel == NULL)
824                                 return -EINVAL;
825                 }
826                 if (!(opt->opt_nflen|opt->opt_flen))
827                         opt = NULL;
828         }
829         if (opt == NULL)
830                 opt = np->opt;
831         if (flowlabel)
832                 opt = fl6_merge_options(&opt_space, flowlabel, opt);
833         opt = ipv6_fixup_options(&opt_space, opt);
834
835         fl.proto = proto;
836         err = rawv6_probe_proto_opt(&fl, msg);
837         if (err)
838                 goto out;
839
840         if (!ipv6_addr_any(daddr))
841                 ipv6_addr_copy(&fl.fl6_dst, daddr);
842         else
843                 fl.fl6_dst.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
844         if (ipv6_addr_any(&fl.fl6_src) && !ipv6_addr_any(&np->saddr))
845                 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
846
847         /* merge ip6_build_xmit from ip6_output */
848         if (opt && opt->srcrt) {
849                 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
850                 ipv6_addr_copy(&final, &fl.fl6_dst);
851                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
852                 final_p = &final;
853         }
854
855         if (!fl.oif && ipv6_addr_is_multicast(&fl.fl6_dst))
856                 fl.oif = np->mcast_oif;
857         security_sk_classify_flow(sk, &fl);
858
859         err = ip6_dst_lookup(sk, &dst, &fl);
860         if (err)
861                 goto out;
862         if (final_p)
863                 ipv6_addr_copy(&fl.fl6_dst, final_p);
864
865         if ((err = __xfrm_lookup(&dst, &fl, sk, XFRM_LOOKUP_WAIT)) < 0) {
866                 if (err == -EREMOTE)
867                         err = ip6_dst_blackhole(sk, &dst, &fl);
868                 if (err < 0)
869                         goto out;
870         }
871
872         if (hlimit < 0) {
873                 if (ipv6_addr_is_multicast(&fl.fl6_dst))
874                         hlimit = np->mcast_hops;
875                 else
876                         hlimit = np->hop_limit;
877                 if (hlimit < 0)
878                         hlimit = ip6_dst_hoplimit(dst);
879         }
880
881         if (tclass < 0) {
882                 tclass = np->tclass;
883                 if (tclass < 0)
884                         tclass = 0;
885         }
886
887         if (msg->msg_flags&MSG_CONFIRM)
888                 goto do_confirm;
889
890 back_from_confirm:
891         if (inet->hdrincl) {
892                 err = rawv6_send_hdrinc(sk, msg->msg_iov, len, &fl, (struct rt6_info*)dst, msg->msg_flags);
893         } else {
894                 lock_sock(sk);
895                 err = ip6_append_data(sk, ip_generic_getfrag, msg->msg_iov,
896                         len, 0, hlimit, tclass, opt, &fl, (struct rt6_info*)dst,
897                         msg->msg_flags);
898
899                 if (err)
900                         ip6_flush_pending_frames(sk);
901                 else if (!(msg->msg_flags & MSG_MORE))
902                         err = rawv6_push_pending_frames(sk, &fl, rp);
903                 release_sock(sk);
904         }
905 done:
906         dst_release(dst);
907 out:
908         fl6_sock_release(flowlabel);
909         return err<0?err:len;
910 do_confirm:
911         dst_confirm(dst);
912         if (!(msg->msg_flags & MSG_PROBE) || len)
913                 goto back_from_confirm;
914         err = 0;
915         goto done;
916 }
917
918 static int rawv6_seticmpfilter(struct sock *sk, int level, int optname,
919                                char __user *optval, int optlen)
920 {
921         switch (optname) {
922         case ICMPV6_FILTER:
923                 if (optlen > sizeof(struct icmp6_filter))
924                         optlen = sizeof(struct icmp6_filter);
925                 if (copy_from_user(&raw6_sk(sk)->filter, optval, optlen))
926                         return -EFAULT;
927                 return 0;
928         default:
929                 return -ENOPROTOOPT;
930         }
931
932         return 0;
933 }
934
935 static int rawv6_geticmpfilter(struct sock *sk, int level, int optname,
936                                char __user *optval, int __user *optlen)
937 {
938         int len;
939
940         switch (optname) {
941         case ICMPV6_FILTER:
942                 if (get_user(len, optlen))
943                         return -EFAULT;
944                 if (len < 0)
945                         return -EINVAL;
946                 if (len > sizeof(struct icmp6_filter))
947                         len = sizeof(struct icmp6_filter);
948                 if (put_user(len, optlen))
949                         return -EFAULT;
950                 if (copy_to_user(optval, &raw6_sk(sk)->filter, len))
951                         return -EFAULT;
952                 return 0;
953         default:
954                 return -ENOPROTOOPT;
955         }
956
957         return 0;
958 }
959
960
961 static int do_rawv6_setsockopt(struct sock *sk, int level, int optname,
962                             char __user *optval, int optlen)
963 {
964         struct raw6_sock *rp = raw6_sk(sk);
965         int val;
966
967         if (get_user(val, (int __user *)optval))
968                 return -EFAULT;
969
970         switch (optname) {
971                 case IPV6_CHECKSUM:
972                         if (inet_sk(sk)->num == IPPROTO_ICMPV6 &&
973                             level == IPPROTO_IPV6) {
974                                 /*
975                                  * RFC3542 tells that IPV6_CHECKSUM socket
976                                  * option in the IPPROTO_IPV6 level is not
977                                  * allowed on ICMPv6 sockets.
978                                  * If you want to set it, use IPPROTO_RAW
979                                  * level IPV6_CHECKSUM socket option
980                                  * (Linux extension).
981                                  */
982                                 return -EINVAL;
983                         }
984
985                         /* You may get strange result with a positive odd offset;
986                            RFC2292bis agrees with me. */
987                         if (val > 0 && (val&1))
988                                 return(-EINVAL);
989                         if (val < 0) {
990                                 rp->checksum = 0;
991                         } else {
992                                 rp->checksum = 1;
993                                 rp->offset = val;
994                         }
995
996                         return 0;
997                         break;
998
999                 default:
1000                         return(-ENOPROTOOPT);
1001         }
1002 }
1003
1004 static int rawv6_setsockopt(struct sock *sk, int level, int optname,
1005                           char __user *optval, int optlen)
1006 {
1007         switch(level) {
1008                 case SOL_RAW:
1009                         break;
1010
1011                 case SOL_ICMPV6:
1012                         if (inet_sk(sk)->num != IPPROTO_ICMPV6)
1013                                 return -EOPNOTSUPP;
1014                         return rawv6_seticmpfilter(sk, level, optname, optval,
1015                                                    optlen);
1016                 case SOL_IPV6:
1017                         if (optname == IPV6_CHECKSUM)
1018                                 break;
1019                 default:
1020                         return ipv6_setsockopt(sk, level, optname, optval,
1021                                                optlen);
1022         }
1023
1024         return do_rawv6_setsockopt(sk, level, optname, optval, optlen);
1025 }
1026
1027 #ifdef CONFIG_COMPAT
1028 static int compat_rawv6_setsockopt(struct sock *sk, int level, int optname,
1029                                    char __user *optval, int optlen)
1030 {
1031         switch (level) {
1032         case SOL_RAW:
1033                 break;
1034         case SOL_ICMPV6:
1035                 if (inet_sk(sk)->num != IPPROTO_ICMPV6)
1036                         return -EOPNOTSUPP;
1037                 return rawv6_seticmpfilter(sk, level, optname, optval, optlen);
1038         case SOL_IPV6:
1039                 if (optname == IPV6_CHECKSUM)
1040                         break;
1041         default:
1042                 return compat_ipv6_setsockopt(sk, level, optname,
1043                                               optval, optlen);
1044         }
1045         return do_rawv6_setsockopt(sk, level, optname, optval, optlen);
1046 }
1047 #endif
1048
1049 static int do_rawv6_getsockopt(struct sock *sk, int level, int optname,
1050                             char __user *optval, int __user *optlen)
1051 {
1052         struct raw6_sock *rp = raw6_sk(sk);
1053         int val, len;
1054
1055         if (get_user(len,optlen))
1056                 return -EFAULT;
1057
1058         switch (optname) {
1059         case IPV6_CHECKSUM:
1060                 /*
1061                  * We allow getsockopt() for IPPROTO_IPV6-level
1062                  * IPV6_CHECKSUM socket option on ICMPv6 sockets
1063                  * since RFC3542 is silent about it.
1064                  */
1065                 if (rp->checksum == 0)
1066                         val = -1;
1067                 else
1068                         val = rp->offset;
1069                 break;
1070
1071         default:
1072                 return -ENOPROTOOPT;
1073         }
1074
1075         len = min_t(unsigned int, sizeof(int), len);
1076
1077         if (put_user(len, optlen))
1078                 return -EFAULT;
1079         if (copy_to_user(optval,&val,len))
1080                 return -EFAULT;
1081         return 0;
1082 }
1083
1084 static int rawv6_getsockopt(struct sock *sk, int level, int optname,
1085                           char __user *optval, int __user *optlen)
1086 {
1087         switch(level) {
1088                 case SOL_RAW:
1089                         break;
1090
1091                 case SOL_ICMPV6:
1092                         if (inet_sk(sk)->num != IPPROTO_ICMPV6)
1093                                 return -EOPNOTSUPP;
1094                         return rawv6_geticmpfilter(sk, level, optname, optval,
1095                                                    optlen);
1096                 case SOL_IPV6:
1097                         if (optname == IPV6_CHECKSUM)
1098                                 break;
1099                 default:
1100                         return ipv6_getsockopt(sk, level, optname, optval,
1101                                                optlen);
1102         }
1103
1104         return do_rawv6_getsockopt(sk, level, optname, optval, optlen);
1105 }
1106
1107 #ifdef CONFIG_COMPAT
1108 static int compat_rawv6_getsockopt(struct sock *sk, int level, int optname,
1109                                    char __user *optval, int __user *optlen)
1110 {
1111         switch (level) {
1112         case SOL_RAW:
1113                 break;
1114         case SOL_ICMPV6:
1115                 if (inet_sk(sk)->num != IPPROTO_ICMPV6)
1116                         return -EOPNOTSUPP;
1117                 return rawv6_geticmpfilter(sk, level, optname, optval, optlen);
1118         case SOL_IPV6:
1119                 if (optname == IPV6_CHECKSUM)
1120                         break;
1121         default:
1122                 return compat_ipv6_getsockopt(sk, level, optname,
1123                                               optval, optlen);
1124         }
1125         return do_rawv6_getsockopt(sk, level, optname, optval, optlen);
1126 }
1127 #endif
1128
1129 static int rawv6_ioctl(struct sock *sk, int cmd, unsigned long arg)
1130 {
1131         switch(cmd) {
1132                 case SIOCOUTQ:
1133                 {
1134                         int amount = atomic_read(&sk->sk_wmem_alloc);
1135                         return put_user(amount, (int __user *)arg);
1136                 }
1137                 case SIOCINQ:
1138                 {
1139                         struct sk_buff *skb;
1140                         int amount = 0;
1141
1142                         spin_lock_bh(&sk->sk_receive_queue.lock);
1143                         skb = skb_peek(&sk->sk_receive_queue);
1144                         if (skb != NULL)
1145                                 amount = skb->tail - skb->transport_header;
1146                         spin_unlock_bh(&sk->sk_receive_queue.lock);
1147                         return put_user(amount, (int __user *)arg);
1148                 }
1149
1150                 default:
1151 #ifdef CONFIG_IPV6_MROUTE
1152                         return ip6mr_ioctl(sk, cmd, (void __user *)arg);
1153 #else
1154                         return -ENOIOCTLCMD;
1155 #endif
1156         }
1157 }
1158
1159 static void rawv6_close(struct sock *sk, long timeout)
1160 {
1161         if (inet_sk(sk)->num == IPPROTO_RAW)
1162                 ip6_ra_control(sk, -1, NULL);
1163         ip6mr_sk_done(sk);
1164         sk_common_release(sk);
1165 }
1166
1167 static int raw6_destroy(struct sock *sk)
1168 {
1169         lock_sock(sk);
1170         ip6_flush_pending_frames(sk);
1171         release_sock(sk);
1172         return 0;
1173 }
1174
1175 static int rawv6_init_sk(struct sock *sk)
1176 {
1177         struct raw6_sock *rp = raw6_sk(sk);
1178
1179         switch (inet_sk(sk)->num) {
1180         case IPPROTO_ICMPV6:
1181                 rp->checksum = 1;
1182                 rp->offset   = 2;
1183                 break;
1184         case IPPROTO_MH:
1185                 rp->checksum = 1;
1186                 rp->offset   = 4;
1187                 break;
1188         default:
1189                 break;
1190         }
1191         return(0);
1192 }
1193
1194 struct proto rawv6_prot = {
1195         .name              = "RAWv6",
1196         .owner             = THIS_MODULE,
1197         .close             = rawv6_close,
1198         .destroy           = raw6_destroy,
1199         .connect           = ip6_datagram_connect,
1200         .disconnect        = udp_disconnect,
1201         .ioctl             = rawv6_ioctl,
1202         .init              = rawv6_init_sk,
1203         .destroy           = inet6_destroy_sock,
1204         .setsockopt        = rawv6_setsockopt,
1205         .getsockopt        = rawv6_getsockopt,
1206         .sendmsg           = rawv6_sendmsg,
1207         .recvmsg           = rawv6_recvmsg,
1208         .bind              = rawv6_bind,
1209         .backlog_rcv       = rawv6_rcv_skb,
1210         .hash              = raw_hash_sk,
1211         .unhash            = raw_unhash_sk,
1212         .obj_size          = sizeof(struct raw6_sock),
1213         .h.raw_hash        = &raw_v6_hashinfo,
1214 #ifdef CONFIG_COMPAT
1215         .compat_setsockopt = compat_rawv6_setsockopt,
1216         .compat_getsockopt = compat_rawv6_getsockopt,
1217 #endif
1218 };
1219
1220 #ifdef CONFIG_PROC_FS
1221 static void raw6_sock_seq_show(struct seq_file *seq, struct sock *sp, int i)
1222 {
1223         struct ipv6_pinfo *np = inet6_sk(sp);
1224         struct in6_addr *dest, *src;
1225         __u16 destp, srcp;
1226
1227         dest  = &np->daddr;
1228         src   = &np->rcv_saddr;
1229         destp = 0;
1230         srcp  = inet_sk(sp)->num;
1231         seq_printf(seq,
1232                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1233                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d\n",
1234                    i,
1235                    src->s6_addr32[0], src->s6_addr32[1],
1236                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1237                    dest->s6_addr32[0], dest->s6_addr32[1],
1238                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1239                    sp->sk_state,
1240                    atomic_read(&sp->sk_wmem_alloc),
1241                    atomic_read(&sp->sk_rmem_alloc),
1242                    0, 0L, 0,
1243                    sock_i_uid(sp), 0,
1244                    sock_i_ino(sp),
1245                    atomic_read(&sp->sk_refcnt), sp, atomic_read(&sp->sk_drops));
1246 }
1247
1248 static int raw6_seq_show(struct seq_file *seq, void *v)
1249 {
1250         if (v == SEQ_START_TOKEN)
1251                 seq_printf(seq,
1252                            "  sl  "
1253                            "local_address                         "
1254                            "remote_address                        "
1255                            "st tx_queue rx_queue tr tm->when retrnsmt"
1256                            "   uid  timeout inode  drops\n");
1257         else
1258                 raw6_sock_seq_show(seq, v, raw_seq_private(seq)->bucket);
1259         return 0;
1260 }
1261
1262 static const struct seq_operations raw6_seq_ops = {
1263         .start =        raw_seq_start,
1264         .next =         raw_seq_next,
1265         .stop =         raw_seq_stop,
1266         .show =         raw6_seq_show,
1267 };
1268
1269 static int raw6_seq_open(struct inode *inode, struct file *file)
1270 {
1271         return raw_seq_open(inode, file, &raw_v6_hashinfo, &raw6_seq_ops);
1272 }
1273
1274 static const struct file_operations raw6_seq_fops = {
1275         .owner =        THIS_MODULE,
1276         .open =         raw6_seq_open,
1277         .read =         seq_read,
1278         .llseek =       seq_lseek,
1279         .release =      seq_release_net,
1280 };
1281
1282 static int raw6_init_net(struct net *net)
1283 {
1284         if (!proc_net_fops_create(net, "raw6", S_IRUGO, &raw6_seq_fops))
1285                 return -ENOMEM;
1286
1287         return 0;
1288 }
1289
1290 static void raw6_exit_net(struct net *net)
1291 {
1292         proc_net_remove(net, "raw6");
1293 }
1294
1295 static struct pernet_operations raw6_net_ops = {
1296         .init = raw6_init_net,
1297         .exit = raw6_exit_net,
1298 };
1299
1300 int __init raw6_proc_init(void)
1301 {
1302         return register_pernet_subsys(&raw6_net_ops);
1303 }
1304
1305 void raw6_proc_exit(void)
1306 {
1307         unregister_pernet_subsys(&raw6_net_ops);
1308 }
1309 #endif  /* CONFIG_PROC_FS */
1310
1311 /* Same as inet6_dgram_ops, sans udp_poll.  */
1312 static const struct proto_ops inet6_sockraw_ops = {
1313         .family            = PF_INET6,
1314         .owner             = THIS_MODULE,
1315         .release           = inet6_release,
1316         .bind              = inet6_bind,
1317         .connect           = inet_dgram_connect,        /* ok           */
1318         .socketpair        = sock_no_socketpair,        /* a do nothing */
1319         .accept            = sock_no_accept,            /* a do nothing */
1320         .getname           = inet6_getname,
1321         .poll              = datagram_poll,             /* ok           */
1322         .ioctl             = inet6_ioctl,               /* must change  */
1323         .listen            = sock_no_listen,            /* ok           */
1324         .shutdown          = inet_shutdown,             /* ok           */
1325         .setsockopt        = sock_common_setsockopt,    /* ok           */
1326         .getsockopt        = sock_common_getsockopt,    /* ok           */
1327         .sendmsg           = inet_sendmsg,              /* ok           */
1328         .recvmsg           = sock_common_recvmsg,       /* ok           */
1329         .mmap              = sock_no_mmap,
1330         .sendpage          = sock_no_sendpage,
1331 #ifdef CONFIG_COMPAT
1332         .compat_setsockopt = compat_sock_common_setsockopt,
1333         .compat_getsockopt = compat_sock_common_getsockopt,
1334 #endif
1335 };
1336
1337 static struct inet_protosw rawv6_protosw = {
1338         .type           = SOCK_RAW,
1339         .protocol       = IPPROTO_IP,   /* wild card */
1340         .prot           = &rawv6_prot,
1341         .ops            = &inet6_sockraw_ops,
1342         .capability     = CAP_NET_RAW,
1343         .no_check       = UDP_CSUM_DEFAULT,
1344         .flags          = INET_PROTOSW_REUSE,
1345 };
1346
1347 int __init rawv6_init(void)
1348 {
1349         int ret;
1350
1351         ret = inet6_register_protosw(&rawv6_protosw);
1352         if (ret)
1353                 goto out;
1354 out:
1355         return ret;
1356 }
1357
1358 void rawv6_exit(void)
1359 {
1360         inet6_unregister_protosw(&rawv6_protosw);
1361 }