xref: /openbmc/linux/net/ipv6/udp.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
1 /*
2  *	UDP over IPv6
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	Based on linux/ipv4/udp.c
9  *
10  *	$Id: udp.c,v 1.65 2002/02/01 22:01:04 davem Exp $
11  *
12  *	Fixes:
13  *	Hideaki YOSHIFUJI	:	sin6_scope_id support
14  *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
15  *	Alexey Kuznetsov		allow both IPv4 and IPv6 sockets to bind
16  *					a single port at the same time.
17  *      Kazunori MIYAZAWA @USAGI:       change process style to use ip6_append_data
18  *      YOSHIFUJI Hideaki @USAGI:	convert /proc/net/udp6 to seq_file.
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 #include <linux/config.h>
27 #include <linux/errno.h>
28 #include <linux/types.h>
29 #include <linux/socket.h>
30 #include <linux/sockios.h>
31 #include <linux/sched.h>
32 #include <linux/net.h>
33 #include <linux/in6.h>
34 #include <linux/netdevice.h>
35 #include <linux/if_arp.h>
36 #include <linux/ipv6.h>
37 #include <linux/icmpv6.h>
38 #include <linux/init.h>
39 #include <asm/uaccess.h>
40 
41 #include <net/sock.h>
42 #include <net/snmp.h>
43 
44 #include <net/ipv6.h>
45 #include <net/ndisc.h>
46 #include <net/protocol.h>
47 #include <net/transp_v6.h>
48 #include <net/ip6_route.h>
49 #include <net/addrconf.h>
50 #include <net/ip.h>
51 #include <net/udp.h>
52 #include <net/raw.h>
53 #include <net/inet_common.h>
54 #include <net/tcp_states.h>
55 
56 #include <net/ip6_checksum.h>
57 #include <net/xfrm.h>
58 
59 #include <linux/proc_fs.h>
60 #include <linux/seq_file.h>
61 
62 DEFINE_SNMP_STAT(struct udp_mib, udp_stats_in6) __read_mostly;
63 
64 /* Grrr, addr_type already calculated by caller, but I don't want
65  * to add some silly "cookie" argument to this method just for that.
66  */
67 static int udp_v6_get_port(struct sock *sk, unsigned short snum)
68 {
69 	struct sock *sk2;
70 	struct hlist_node *node;
71 
72 	write_lock_bh(&udp_hash_lock);
73 	if (snum == 0) {
74 		int best_size_so_far, best, result, i;
75 
76 		if (udp_port_rover > sysctl_local_port_range[1] ||
77 		    udp_port_rover < sysctl_local_port_range[0])
78 			udp_port_rover = sysctl_local_port_range[0];
79 		best_size_so_far = 32767;
80 		best = result = udp_port_rover;
81 		for (i = 0; i < UDP_HTABLE_SIZE; i++, result++) {
82 			int size;
83 			struct hlist_head *list;
84 
85 			list = &udp_hash[result & (UDP_HTABLE_SIZE - 1)];
86 			if (hlist_empty(list)) {
87 				if (result > sysctl_local_port_range[1])
88 					result = sysctl_local_port_range[0] +
89 						((result - sysctl_local_port_range[0]) &
90 						 (UDP_HTABLE_SIZE - 1));
91 				goto gotit;
92 			}
93 			size = 0;
94 			sk_for_each(sk2, node, list)
95 				if (++size >= best_size_so_far)
96 					goto next;
97 			best_size_so_far = size;
98 			best = result;
99 		next:;
100 		}
101 		result = best;
102 		for(i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++, result += UDP_HTABLE_SIZE) {
103 			if (result > sysctl_local_port_range[1])
104 				result = sysctl_local_port_range[0]
105 					+ ((result - sysctl_local_port_range[0]) &
106 					   (UDP_HTABLE_SIZE - 1));
107 			if (!udp_lport_inuse(result))
108 				break;
109 		}
110 		if (i >= (1 << 16) / UDP_HTABLE_SIZE)
111 			goto fail;
112 gotit:
113 		udp_port_rover = snum = result;
114 	} else {
115 		sk_for_each(sk2, node,
116 			    &udp_hash[snum & (UDP_HTABLE_SIZE - 1)]) {
117 			if (inet_sk(sk2)->num == snum &&
118 			    sk2 != sk &&
119 			    (!sk2->sk_bound_dev_if ||
120 			     !sk->sk_bound_dev_if ||
121 			     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
122 			    (!sk2->sk_reuse || !sk->sk_reuse) &&
123 			    ipv6_rcv_saddr_equal(sk, sk2))
124 				goto fail;
125 		}
126 	}
127 
128 	inet_sk(sk)->num = snum;
129 	if (sk_unhashed(sk)) {
130 		sk_add_node(sk, &udp_hash[snum & (UDP_HTABLE_SIZE - 1)]);
131 		sock_prot_inc_use(sk->sk_prot);
132 	}
133 	write_unlock_bh(&udp_hash_lock);
134 	return 0;
135 
136 fail:
137 	write_unlock_bh(&udp_hash_lock);
138 	return 1;
139 }
140 
141 static void udp_v6_hash(struct sock *sk)
142 {
143 	BUG();
144 }
145 
146 static void udp_v6_unhash(struct sock *sk)
147 {
148  	write_lock_bh(&udp_hash_lock);
149 	if (sk_del_node_init(sk)) {
150 		inet_sk(sk)->num = 0;
151 		sock_prot_dec_use(sk->sk_prot);
152 	}
153 	write_unlock_bh(&udp_hash_lock);
154 }
155 
156 static struct sock *udp_v6_lookup(struct in6_addr *saddr, u16 sport,
157 				  struct in6_addr *daddr, u16 dport, int dif)
158 {
159 	struct sock *sk, *result = NULL;
160 	struct hlist_node *node;
161 	unsigned short hnum = ntohs(dport);
162 	int badness = -1;
163 
164  	read_lock(&udp_hash_lock);
165 	sk_for_each(sk, node, &udp_hash[hnum & (UDP_HTABLE_SIZE - 1)]) {
166 		struct inet_sock *inet = inet_sk(sk);
167 
168 		if (inet->num == hnum && sk->sk_family == PF_INET6) {
169 			struct ipv6_pinfo *np = inet6_sk(sk);
170 			int score = 0;
171 			if (inet->dport) {
172 				if (inet->dport != sport)
173 					continue;
174 				score++;
175 			}
176 			if (!ipv6_addr_any(&np->rcv_saddr)) {
177 				if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
178 					continue;
179 				score++;
180 			}
181 			if (!ipv6_addr_any(&np->daddr)) {
182 				if (!ipv6_addr_equal(&np->daddr, saddr))
183 					continue;
184 				score++;
185 			}
186 			if (sk->sk_bound_dev_if) {
187 				if (sk->sk_bound_dev_if != dif)
188 					continue;
189 				score++;
190 			}
191 			if(score == 4) {
192 				result = sk;
193 				break;
194 			} else if(score > badness) {
195 				result = sk;
196 				badness = score;
197 			}
198 		}
199 	}
200 	if (result)
201 		sock_hold(result);
202  	read_unlock(&udp_hash_lock);
203 	return result;
204 }
205 
206 /*
207  *
208  */
209 
210 static void udpv6_close(struct sock *sk, long timeout)
211 {
212 	sk_common_release(sk);
213 }
214 
215 /*
216  * 	This should be easy, if there is something there we
217  * 	return it, otherwise we block.
218  */
219 
220 static int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
221 		  struct msghdr *msg, size_t len,
222 		  int noblock, int flags, int *addr_len)
223 {
224 	struct ipv6_pinfo *np = inet6_sk(sk);
225 	struct inet_sock *inet = inet_sk(sk);
226   	struct sk_buff *skb;
227 	size_t copied;
228   	int err;
229 
230   	if (addr_len)
231   		*addr_len=sizeof(struct sockaddr_in6);
232 
233 	if (flags & MSG_ERRQUEUE)
234 		return ipv6_recv_error(sk, msg, len);
235 
236 try_again:
237 	skb = skb_recv_datagram(sk, flags, noblock, &err);
238 	if (!skb)
239 		goto out;
240 
241  	copied = skb->len - sizeof(struct udphdr);
242   	if (copied > len) {
243   		copied = len;
244   		msg->msg_flags |= MSG_TRUNC;
245   	}
246 
247 	if (skb->ip_summed==CHECKSUM_UNNECESSARY) {
248 		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov,
249 					      copied);
250 	} else if (msg->msg_flags&MSG_TRUNC) {
251 		if ((unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum)))
252 			goto csum_copy_err;
253 		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov,
254 					      copied);
255 	} else {
256 		err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
257 		if (err == -EINVAL)
258 			goto csum_copy_err;
259 	}
260 	if (err)
261 		goto out_free;
262 
263 	sock_recv_timestamp(msg, sk, skb);
264 
265 	/* Copy the address. */
266 	if (msg->msg_name) {
267 		struct sockaddr_in6 *sin6;
268 
269 		sin6 = (struct sockaddr_in6 *) msg->msg_name;
270 		sin6->sin6_family = AF_INET6;
271 		sin6->sin6_port = skb->h.uh->source;
272 		sin6->sin6_flowinfo = 0;
273 		sin6->sin6_scope_id = 0;
274 
275 		if (skb->protocol == htons(ETH_P_IP))
276 			ipv6_addr_set(&sin6->sin6_addr, 0, 0,
277 				      htonl(0xffff), skb->nh.iph->saddr);
278 		else {
279 			ipv6_addr_copy(&sin6->sin6_addr, &skb->nh.ipv6h->saddr);
280 			if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
281 				sin6->sin6_scope_id = IP6CB(skb)->iif;
282 		}
283 
284 	}
285 	if (skb->protocol == htons(ETH_P_IP)) {
286 		if (inet->cmsg_flags)
287 			ip_cmsg_recv(msg, skb);
288 	} else {
289 		if (np->rxopt.all)
290 			datagram_recv_ctl(sk, msg, skb);
291   	}
292 
293 	err = copied;
294 	if (flags & MSG_TRUNC)
295 		err = skb->len - sizeof(struct udphdr);
296 
297 out_free:
298 	skb_free_datagram(sk, skb);
299 out:
300 	return err;
301 
302 csum_copy_err:
303 	/* Clear queue. */
304 	if (flags&MSG_PEEK) {
305 		int clear = 0;
306 		spin_lock_bh(&sk->sk_receive_queue.lock);
307 		if (skb == skb_peek(&sk->sk_receive_queue)) {
308 			__skb_unlink(skb, &sk->sk_receive_queue);
309 			clear = 1;
310 		}
311 		spin_unlock_bh(&sk->sk_receive_queue.lock);
312 		if (clear)
313 			kfree_skb(skb);
314 	}
315 
316 	skb_free_datagram(sk, skb);
317 
318 	if (flags & MSG_DONTWAIT) {
319 		UDP6_INC_STATS_USER(UDP_MIB_INERRORS);
320 		return -EAGAIN;
321 	}
322 	goto try_again;
323 }
324 
325 static void udpv6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
326 	       int type, int code, int offset, __u32 info)
327 {
328 	struct ipv6_pinfo *np;
329 	struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
330 	struct net_device *dev = skb->dev;
331 	struct in6_addr *saddr = &hdr->saddr;
332 	struct in6_addr *daddr = &hdr->daddr;
333 	struct udphdr *uh = (struct udphdr*)(skb->data+offset);
334 	struct sock *sk;
335 	int err;
336 
337 	sk = udp_v6_lookup(daddr, uh->dest, saddr, uh->source, dev->ifindex);
338 
339 	if (sk == NULL)
340 		return;
341 
342 	np = inet6_sk(sk);
343 
344 	if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
345 		goto out;
346 
347 	if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
348 		goto out;
349 
350 	if (np->recverr)
351 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
352 
353 	sk->sk_err = err;
354 	sk->sk_error_report(sk);
355 out:
356 	sock_put(sk);
357 }
358 
359 static inline int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
360 {
361 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
362 		kfree_skb(skb);
363 		return -1;
364 	}
365 
366 	if (skb->ip_summed != CHECKSUM_UNNECESSARY) {
367 		if ((unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum))) {
368 			UDP6_INC_STATS_BH(UDP_MIB_INERRORS);
369 			kfree_skb(skb);
370 			return 0;
371 		}
372 		skb->ip_summed = CHECKSUM_UNNECESSARY;
373 	}
374 
375 	if (sock_queue_rcv_skb(sk,skb)<0) {
376 		UDP6_INC_STATS_BH(UDP_MIB_INERRORS);
377 		kfree_skb(skb);
378 		return 0;
379 	}
380 	UDP6_INC_STATS_BH(UDP_MIB_INDATAGRAMS);
381 	return 0;
382 }
383 
384 static struct sock *udp_v6_mcast_next(struct sock *sk,
385 				      u16 loc_port, struct in6_addr *loc_addr,
386 				      u16 rmt_port, struct in6_addr *rmt_addr,
387 				      int dif)
388 {
389 	struct hlist_node *node;
390 	struct sock *s = sk;
391 	unsigned short num = ntohs(loc_port);
392 
393 	sk_for_each_from(s, node) {
394 		struct inet_sock *inet = inet_sk(s);
395 
396 		if (inet->num == num && s->sk_family == PF_INET6) {
397 			struct ipv6_pinfo *np = inet6_sk(s);
398 			if (inet->dport) {
399 				if (inet->dport != rmt_port)
400 					continue;
401 			}
402 			if (!ipv6_addr_any(&np->daddr) &&
403 			    !ipv6_addr_equal(&np->daddr, rmt_addr))
404 				continue;
405 
406 			if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
407 				continue;
408 
409 			if (!ipv6_addr_any(&np->rcv_saddr)) {
410 				if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
411 					continue;
412 			}
413 			if(!inet6_mc_check(s, loc_addr, rmt_addr))
414 				continue;
415 			return s;
416 		}
417 	}
418 	return NULL;
419 }
420 
421 /*
422  * Note: called only from the BH handler context,
423  * so we don't need to lock the hashes.
424  */
425 static void udpv6_mcast_deliver(struct udphdr *uh,
426 				struct in6_addr *saddr, struct in6_addr *daddr,
427 				struct sk_buff *skb)
428 {
429 	struct sock *sk, *sk2;
430 	int dif;
431 
432 	read_lock(&udp_hash_lock);
433 	sk = sk_head(&udp_hash[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
434 	dif = skb->dev->ifindex;
435 	sk = udp_v6_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
436 	if (!sk) {
437 		kfree_skb(skb);
438 		goto out;
439 	}
440 
441 	sk2 = sk;
442 	while ((sk2 = udp_v6_mcast_next(sk_next(sk2), uh->dest, daddr,
443 					uh->source, saddr, dif))) {
444 		struct sk_buff *buff = skb_clone(skb, GFP_ATOMIC);
445 		if (buff)
446 			udpv6_queue_rcv_skb(sk2, buff);
447 	}
448 	udpv6_queue_rcv_skb(sk, skb);
449 out:
450 	read_unlock(&udp_hash_lock);
451 }
452 
453 static int udpv6_rcv(struct sk_buff **pskb, unsigned int *nhoffp)
454 {
455 	struct sk_buff *skb = *pskb;
456 	struct sock *sk;
457   	struct udphdr *uh;
458 	struct net_device *dev = skb->dev;
459 	struct in6_addr *saddr, *daddr;
460 	u32 ulen = 0;
461 
462 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
463 		goto short_packet;
464 
465 	saddr = &skb->nh.ipv6h->saddr;
466 	daddr = &skb->nh.ipv6h->daddr;
467 	uh = skb->h.uh;
468 
469 	ulen = ntohs(uh->len);
470 
471 	/* Check for jumbo payload */
472 	if (ulen == 0)
473 		ulen = skb->len;
474 
475 	if (ulen > skb->len || ulen < sizeof(*uh))
476 		goto short_packet;
477 
478 	if (uh->check == 0) {
479 		/* RFC 2460 section 8.1 says that we SHOULD log
480 		   this error. Well, it is reasonable.
481 		 */
482 		LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
483 		goto discard;
484 	}
485 
486 	if (ulen < skb->len) {
487 		if (pskb_trim_rcsum(skb, ulen))
488 			goto discard;
489 		saddr = &skb->nh.ipv6h->saddr;
490 		daddr = &skb->nh.ipv6h->daddr;
491 		uh = skb->h.uh;
492 	}
493 
494 	if (skb->ip_summed==CHECKSUM_HW) {
495 		skb->ip_summed = CHECKSUM_UNNECESSARY;
496 		if (csum_ipv6_magic(saddr, daddr, ulen, IPPROTO_UDP, skb->csum)) {
497 			LIMIT_NETDEBUG(KERN_DEBUG "udp v6 hw csum failure.\n");
498 			skb->ip_summed = CHECKSUM_NONE;
499 		}
500 	}
501 	if (skb->ip_summed != CHECKSUM_UNNECESSARY)
502 		skb->csum = ~csum_ipv6_magic(saddr, daddr, ulen, IPPROTO_UDP, 0);
503 
504 	/*
505 	 *	Multicast receive code
506 	 */
507 	if (ipv6_addr_is_multicast(daddr)) {
508 		udpv6_mcast_deliver(uh, saddr, daddr, skb);
509 		return 0;
510 	}
511 
512 	/* Unicast */
513 
514 	/*
515 	 * check socket cache ... must talk to Alan about his plans
516 	 * for sock caches... i'll skip this for now.
517 	 */
518 	sk = udp_v6_lookup(saddr, uh->source, daddr, uh->dest, dev->ifindex);
519 
520 	if (sk == NULL) {
521 		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
522 			goto discard;
523 
524 		if (skb->ip_summed != CHECKSUM_UNNECESSARY &&
525 		    (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum)))
526 			goto discard;
527 		UDP6_INC_STATS_BH(UDP_MIB_NOPORTS);
528 
529 		icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0, dev);
530 
531 		kfree_skb(skb);
532 		return(0);
533 	}
534 
535 	/* deliver */
536 
537 	udpv6_queue_rcv_skb(sk, skb);
538 	sock_put(sk);
539 	return(0);
540 
541 short_packet:
542 	if (net_ratelimit())
543 		printk(KERN_DEBUG "UDP: short packet: %d/%u\n", ulen, skb->len);
544 
545 discard:
546 	UDP6_INC_STATS_BH(UDP_MIB_INERRORS);
547 	kfree_skb(skb);
548 	return(0);
549 }
550 /*
551  * Throw away all pending data and cancel the corking. Socket is locked.
552  */
553 static void udp_v6_flush_pending_frames(struct sock *sk)
554 {
555 	struct udp_sock *up = udp_sk(sk);
556 
557 	if (up->pending) {
558 		up->len = 0;
559 		up->pending = 0;
560 		ip6_flush_pending_frames(sk);
561         }
562 }
563 
564 /*
565  *	Sending
566  */
567 
568 static int udp_v6_push_pending_frames(struct sock *sk, struct udp_sock *up)
569 {
570 	struct sk_buff *skb;
571 	struct udphdr *uh;
572 	struct inet_sock *inet = inet_sk(sk);
573 	struct flowi *fl = &inet->cork.fl;
574 	int err = 0;
575 
576 	/* Grab the skbuff where UDP header space exists. */
577 	if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
578 		goto out;
579 
580 	/*
581 	 * Create a UDP header
582 	 */
583 	uh = skb->h.uh;
584 	uh->source = fl->fl_ip_sport;
585 	uh->dest = fl->fl_ip_dport;
586 	uh->len = htons(up->len);
587 	uh->check = 0;
588 
589 	if (sk->sk_no_check == UDP_CSUM_NOXMIT) {
590 		skb->ip_summed = CHECKSUM_NONE;
591 		goto send;
592 	}
593 
594 	if (skb_queue_len(&sk->sk_write_queue) == 1) {
595 		skb->csum = csum_partial((char *)uh,
596 				sizeof(struct udphdr), skb->csum);
597 		uh->check = csum_ipv6_magic(&fl->fl6_src,
598 					    &fl->fl6_dst,
599 					    up->len, fl->proto, skb->csum);
600 	} else {
601 		u32 tmp_csum = 0;
602 
603 		skb_queue_walk(&sk->sk_write_queue, skb) {
604 			tmp_csum = csum_add(tmp_csum, skb->csum);
605 		}
606 		tmp_csum = csum_partial((char *)uh,
607 				sizeof(struct udphdr), tmp_csum);
608                 tmp_csum = csum_ipv6_magic(&fl->fl6_src,
609 					   &fl->fl6_dst,
610 					   up->len, fl->proto, tmp_csum);
611                 uh->check = tmp_csum;
612 
613 	}
614 	if (uh->check == 0)
615 		uh->check = -1;
616 
617 send:
618 	err = ip6_push_pending_frames(sk);
619 out:
620 	up->len = 0;
621 	up->pending = 0;
622 	return err;
623 }
624 
625 static int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
626 		  struct msghdr *msg, size_t len)
627 {
628 	struct ipv6_txoptions opt_space;
629 	struct udp_sock *up = udp_sk(sk);
630 	struct inet_sock *inet = inet_sk(sk);
631 	struct ipv6_pinfo *np = inet6_sk(sk);
632 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
633 	struct in6_addr *daddr, *final_p = NULL, final;
634 	struct ipv6_txoptions *opt = NULL;
635 	struct ip6_flowlabel *flowlabel = NULL;
636 	struct flowi *fl = &inet->cork.fl;
637 	struct dst_entry *dst;
638 	int addr_len = msg->msg_namelen;
639 	int ulen = len;
640 	int hlimit = -1;
641 	int tclass = -1;
642 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
643 	int err;
644 	int connected = 0;
645 
646 	/* destination address check */
647 	if (sin6) {
648 		if (addr_len < offsetof(struct sockaddr, sa_data))
649 			return -EINVAL;
650 
651 		switch (sin6->sin6_family) {
652 		case AF_INET6:
653 			if (addr_len < SIN6_LEN_RFC2133)
654 				return -EINVAL;
655 			daddr = &sin6->sin6_addr;
656 			break;
657 		case AF_INET:
658 			goto do_udp_sendmsg;
659 		case AF_UNSPEC:
660 			msg->msg_name = sin6 = NULL;
661 			msg->msg_namelen = addr_len = 0;
662 			daddr = NULL;
663 			break;
664 		default:
665 			return -EINVAL;
666 		}
667 	} else if (!up->pending) {
668 		if (sk->sk_state != TCP_ESTABLISHED)
669 			return -EDESTADDRREQ;
670 		daddr = &np->daddr;
671 	} else
672 		daddr = NULL;
673 
674 	if (daddr) {
675 		if (ipv6_addr_type(daddr) == IPV6_ADDR_MAPPED) {
676 			struct sockaddr_in sin;
677 			sin.sin_family = AF_INET;
678 			sin.sin_port = sin6 ? sin6->sin6_port : inet->dport;
679 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
680 			msg->msg_name = &sin;
681 			msg->msg_namelen = sizeof(sin);
682 do_udp_sendmsg:
683 			if (__ipv6_only_sock(sk))
684 				return -ENETUNREACH;
685 			return udp_sendmsg(iocb, sk, msg, len);
686 		}
687 	}
688 
689 	if (up->pending == AF_INET)
690 		return udp_sendmsg(iocb, sk, msg, len);
691 
692 	/* Rough check on arithmetic overflow,
693 	   better check is made in ip6_build_xmit
694 	   */
695 	if (len > INT_MAX - sizeof(struct udphdr))
696 		return -EMSGSIZE;
697 
698 	if (up->pending) {
699 		/*
700 		 * There are pending frames.
701 		 * The socket lock must be held while it's corked.
702 		 */
703 		lock_sock(sk);
704 		if (likely(up->pending)) {
705 			if (unlikely(up->pending != AF_INET6)) {
706 				release_sock(sk);
707 				return -EAFNOSUPPORT;
708 			}
709 			dst = NULL;
710 			goto do_append_data;
711 		}
712 		release_sock(sk);
713 	}
714 	ulen += sizeof(struct udphdr);
715 
716 	memset(fl, 0, sizeof(*fl));
717 
718 	if (sin6) {
719 		if (sin6->sin6_port == 0)
720 			return -EINVAL;
721 
722 		fl->fl_ip_dport = sin6->sin6_port;
723 		daddr = &sin6->sin6_addr;
724 
725 		if (np->sndflow) {
726 			fl->fl6_flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
727 			if (fl->fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
728 				flowlabel = fl6_sock_lookup(sk, fl->fl6_flowlabel);
729 				if (flowlabel == NULL)
730 					return -EINVAL;
731 				daddr = &flowlabel->dst;
732 			}
733 		}
734 
735 		/*
736 		 * Otherwise it will be difficult to maintain
737 		 * sk->sk_dst_cache.
738 		 */
739 		if (sk->sk_state == TCP_ESTABLISHED &&
740 		    ipv6_addr_equal(daddr, &np->daddr))
741 			daddr = &np->daddr;
742 
743 		if (addr_len >= sizeof(struct sockaddr_in6) &&
744 		    sin6->sin6_scope_id &&
745 		    ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
746 			fl->oif = sin6->sin6_scope_id;
747 	} else {
748 		if (sk->sk_state != TCP_ESTABLISHED)
749 			return -EDESTADDRREQ;
750 
751 		fl->fl_ip_dport = inet->dport;
752 		daddr = &np->daddr;
753 		fl->fl6_flowlabel = np->flow_label;
754 		connected = 1;
755 	}
756 
757 	if (!fl->oif)
758 		fl->oif = sk->sk_bound_dev_if;
759 
760 	if (msg->msg_controllen) {
761 		opt = &opt_space;
762 		memset(opt, 0, sizeof(struct ipv6_txoptions));
763 		opt->tot_len = sizeof(*opt);
764 
765 		err = datagram_send_ctl(msg, fl, opt, &hlimit, &tclass);
766 		if (err < 0) {
767 			fl6_sock_release(flowlabel);
768 			return err;
769 		}
770 		if ((fl->fl6_flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
771 			flowlabel = fl6_sock_lookup(sk, fl->fl6_flowlabel);
772 			if (flowlabel == NULL)
773 				return -EINVAL;
774 		}
775 		if (!(opt->opt_nflen|opt->opt_flen))
776 			opt = NULL;
777 		connected = 0;
778 	}
779 	if (opt == NULL)
780 		opt = np->opt;
781 	opt = fl6_merge_options(&opt_space, flowlabel, opt);
782 
783 	fl->proto = IPPROTO_UDP;
784 	ipv6_addr_copy(&fl->fl6_dst, daddr);
785 	if (ipv6_addr_any(&fl->fl6_src) && !ipv6_addr_any(&np->saddr))
786 		ipv6_addr_copy(&fl->fl6_src, &np->saddr);
787 	fl->fl_ip_sport = inet->sport;
788 
789 	/* merge ip6_build_xmit from ip6_output */
790 	if (opt && opt->srcrt) {
791 		struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
792 		ipv6_addr_copy(&final, &fl->fl6_dst);
793 		ipv6_addr_copy(&fl->fl6_dst, rt0->addr);
794 		final_p = &final;
795 		connected = 0;
796 	}
797 
798 	if (!fl->oif && ipv6_addr_is_multicast(&fl->fl6_dst)) {
799 		fl->oif = np->mcast_oif;
800 		connected = 0;
801 	}
802 
803 	err = ip6_dst_lookup(sk, &dst, fl);
804 	if (err)
805 		goto out;
806 	if (final_p)
807 		ipv6_addr_copy(&fl->fl6_dst, final_p);
808 
809 	if ((err = xfrm_lookup(&dst, fl, sk, 0)) < 0)
810 		goto out;
811 
812 	if (hlimit < 0) {
813 		if (ipv6_addr_is_multicast(&fl->fl6_dst))
814 			hlimit = np->mcast_hops;
815 		else
816 			hlimit = np->hop_limit;
817 		if (hlimit < 0)
818 			hlimit = dst_metric(dst, RTAX_HOPLIMIT);
819 		if (hlimit < 0)
820 			hlimit = ipv6_get_hoplimit(dst->dev);
821 	}
822 
823 	if (tclass < 0) {
824 		tclass = np->tclass;
825 		if (tclass < 0)
826 			tclass = 0;
827 	}
828 
829 	if (msg->msg_flags&MSG_CONFIRM)
830 		goto do_confirm;
831 back_from_confirm:
832 
833 	lock_sock(sk);
834 	if (unlikely(up->pending)) {
835 		/* The socket is already corked while preparing it. */
836 		/* ... which is an evident application bug. --ANK */
837 		release_sock(sk);
838 
839 		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
840 		err = -EINVAL;
841 		goto out;
842 	}
843 
844 	up->pending = AF_INET6;
845 
846 do_append_data:
847 	up->len += ulen;
848 	err = ip6_append_data(sk, ip_generic_getfrag, msg->msg_iov, ulen,
849 		sizeof(struct udphdr), hlimit, tclass, opt, fl,
850 		(struct rt6_info*)dst,
851 		corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
852 	if (err)
853 		udp_v6_flush_pending_frames(sk);
854 	else if (!corkreq)
855 		err = udp_v6_push_pending_frames(sk, up);
856 
857 	if (dst) {
858 		if (connected) {
859 			ip6_dst_store(sk, dst,
860 				      ipv6_addr_equal(&fl->fl6_dst, &np->daddr) ?
861 				      &np->daddr : NULL);
862 		} else {
863 			dst_release(dst);
864 		}
865 	}
866 
867 	if (err > 0)
868 		err = np->recverr ? net_xmit_errno(err) : 0;
869 	release_sock(sk);
870 out:
871 	fl6_sock_release(flowlabel);
872 	if (!err) {
873 		UDP6_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS);
874 		return len;
875 	}
876 	return err;
877 
878 do_confirm:
879 	dst_confirm(dst);
880 	if (!(msg->msg_flags&MSG_PROBE) || len)
881 		goto back_from_confirm;
882 	err = 0;
883 	goto out;
884 }
885 
886 static int udpv6_destroy_sock(struct sock *sk)
887 {
888 	lock_sock(sk);
889 	udp_v6_flush_pending_frames(sk);
890 	release_sock(sk);
891 
892 	inet6_destroy_sock(sk);
893 
894 	return 0;
895 }
896 
897 /*
898  *	Socket option code for UDP
899  */
900 static int udpv6_setsockopt(struct sock *sk, int level, int optname,
901 			  char __user *optval, int optlen)
902 {
903 	struct udp_sock *up = udp_sk(sk);
904 	int val;
905 	int err = 0;
906 
907 	if (level != SOL_UDP)
908 		return ipv6_setsockopt(sk, level, optname, optval, optlen);
909 
910 	if(optlen<sizeof(int))
911 		return -EINVAL;
912 
913 	if (get_user(val, (int __user *)optval))
914 		return -EFAULT;
915 
916 	switch(optname) {
917 	case UDP_CORK:
918 		if (val != 0) {
919 			up->corkflag = 1;
920 		} else {
921 			up->corkflag = 0;
922 			lock_sock(sk);
923 			udp_v6_push_pending_frames(sk, up);
924 			release_sock(sk);
925 		}
926 		break;
927 
928 	case UDP_ENCAP:
929 		switch (val) {
930 		case 0:
931 			up->encap_type = val;
932 			break;
933 		default:
934 			err = -ENOPROTOOPT;
935 			break;
936 		}
937 		break;
938 
939 	default:
940 		err = -ENOPROTOOPT;
941 		break;
942 	};
943 
944 	return err;
945 }
946 
947 static int udpv6_getsockopt(struct sock *sk, int level, int optname,
948 			  char __user *optval, int __user *optlen)
949 {
950 	struct udp_sock *up = udp_sk(sk);
951 	int val, len;
952 
953 	if (level != SOL_UDP)
954 		return ipv6_getsockopt(sk, level, optname, optval, optlen);
955 
956 	if(get_user(len,optlen))
957 		return -EFAULT;
958 
959 	len = min_t(unsigned int, len, sizeof(int));
960 
961 	if(len < 0)
962 		return -EINVAL;
963 
964 	switch(optname) {
965 	case UDP_CORK:
966 		val = up->corkflag;
967 		break;
968 
969 	case UDP_ENCAP:
970 		val = up->encap_type;
971 		break;
972 
973 	default:
974 		return -ENOPROTOOPT;
975 	};
976 
977   	if(put_user(len, optlen))
978   		return -EFAULT;
979 	if(copy_to_user(optval, &val,len))
980 		return -EFAULT;
981   	return 0;
982 }
983 
984 static struct inet6_protocol udpv6_protocol = {
985 	.handler	=	udpv6_rcv,
986 	.err_handler	=	udpv6_err,
987 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
988 };
989 
990 /* ------------------------------------------------------------------------ */
991 #ifdef CONFIG_PROC_FS
992 
993 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
994 {
995 	struct inet_sock *inet = inet_sk(sp);
996 	struct ipv6_pinfo *np = inet6_sk(sp);
997 	struct in6_addr *dest, *src;
998 	__u16 destp, srcp;
999 
1000 	dest  = &np->daddr;
1001 	src   = &np->rcv_saddr;
1002 	destp = ntohs(inet->dport);
1003 	srcp  = ntohs(inet->sport);
1004 	seq_printf(seq,
1005 		   "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1006 		   "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p\n",
1007 		   bucket,
1008 		   src->s6_addr32[0], src->s6_addr32[1],
1009 		   src->s6_addr32[2], src->s6_addr32[3], srcp,
1010 		   dest->s6_addr32[0], dest->s6_addr32[1],
1011 		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
1012 		   sp->sk_state,
1013 		   atomic_read(&sp->sk_wmem_alloc),
1014 		   atomic_read(&sp->sk_rmem_alloc),
1015 		   0, 0L, 0,
1016 		   sock_i_uid(sp), 0,
1017 		   sock_i_ino(sp),
1018 		   atomic_read(&sp->sk_refcnt), sp);
1019 }
1020 
1021 static int udp6_seq_show(struct seq_file *seq, void *v)
1022 {
1023 	if (v == SEQ_START_TOKEN)
1024 		seq_printf(seq,
1025 			   "  sl  "
1026 			   "local_address                         "
1027 			   "remote_address                        "
1028 			   "st tx_queue rx_queue tr tm->when retrnsmt"
1029 			   "   uid  timeout inode\n");
1030 	else
1031 		udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1032 	return 0;
1033 }
1034 
1035 static struct file_operations udp6_seq_fops;
1036 static struct udp_seq_afinfo udp6_seq_afinfo = {
1037 	.owner		= THIS_MODULE,
1038 	.name		= "udp6",
1039 	.family		= AF_INET6,
1040 	.seq_show	= udp6_seq_show,
1041 	.seq_fops	= &udp6_seq_fops,
1042 };
1043 
1044 int __init udp6_proc_init(void)
1045 {
1046 	return udp_proc_register(&udp6_seq_afinfo);
1047 }
1048 
1049 void udp6_proc_exit(void) {
1050 	udp_proc_unregister(&udp6_seq_afinfo);
1051 }
1052 #endif /* CONFIG_PROC_FS */
1053 
1054 /* ------------------------------------------------------------------------ */
1055 
1056 struct proto udpv6_prot = {
1057 	.name =		"UDPv6",
1058 	.owner =	THIS_MODULE,
1059 	.close =	udpv6_close,
1060 	.connect =	ip6_datagram_connect,
1061 	.disconnect =	udp_disconnect,
1062 	.ioctl =	udp_ioctl,
1063 	.destroy =	udpv6_destroy_sock,
1064 	.setsockopt =	udpv6_setsockopt,
1065 	.getsockopt =	udpv6_getsockopt,
1066 	.sendmsg =	udpv6_sendmsg,
1067 	.recvmsg =	udpv6_recvmsg,
1068 	.backlog_rcv =	udpv6_queue_rcv_skb,
1069 	.hash =		udp_v6_hash,
1070 	.unhash =	udp_v6_unhash,
1071 	.get_port =	udp_v6_get_port,
1072 	.obj_size =	sizeof(struct udp6_sock),
1073 };
1074 
1075 static struct inet_protosw udpv6_protosw = {
1076 	.type =      SOCK_DGRAM,
1077 	.protocol =  IPPROTO_UDP,
1078 	.prot =      &udpv6_prot,
1079 	.ops =       &inet6_dgram_ops,
1080 	.capability =-1,
1081 	.no_check =  UDP_CSUM_DEFAULT,
1082 	.flags =     INET_PROTOSW_PERMANENT,
1083 };
1084 
1085 
1086 void __init udpv6_init(void)
1087 {
1088 	if (inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP) < 0)
1089 		printk(KERN_ERR "udpv6_init: Could not register protocol\n");
1090 	inet6_register_protosw(&udpv6_protosw);
1091 }
1092