xref: /linux/net/ipv6/udp.c (revision 367b8112fe2ea5c39a7bb4d263dcdd9b612fae18)
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  *	Fixes:
11  *	Hideaki YOSHIFUJI	:	sin6_scope_id support
12  *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
13  *	Alexey Kuznetsov		allow both IPv4 and IPv6 sockets to bind
14  *					a single port at the same time.
15  *      Kazunori MIYAZAWA @USAGI:       change process style to use ip6_append_data
16  *      YOSHIFUJI Hideaki @USAGI:	convert /proc/net/udp6 to seq_file.
17  *
18  *	This program is free software; you can redistribute it and/or
19  *      modify it under the terms of the GNU General Public License
20  *      as published by the Free Software Foundation; either version
21  *      2 of the License, or (at your option) any later version.
22  */
23 
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/socket.h>
27 #include <linux/sockios.h>
28 #include <linux/net.h>
29 #include <linux/in6.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/ipv6.h>
33 #include <linux/icmpv6.h>
34 #include <linux/init.h>
35 #include <linux/module.h>
36 #include <linux/skbuff.h>
37 #include <asm/uaccess.h>
38 
39 #include <net/ndisc.h>
40 #include <net/protocol.h>
41 #include <net/transp_v6.h>
42 #include <net/ip6_route.h>
43 #include <net/raw.h>
44 #include <net/tcp_states.h>
45 #include <net/ip6_checksum.h>
46 #include <net/xfrm.h>
47 
48 #include <linux/proc_fs.h>
49 #include <linux/seq_file.h>
50 #include "udp_impl.h"
51 
52 int udp_v6_get_port(struct sock *sk, unsigned short snum)
53 {
54 	return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal);
55 }
56 
57 static struct sock *__udp6_lib_lookup(struct net *net,
58 				      struct in6_addr *saddr, __be16 sport,
59 				      struct in6_addr *daddr, __be16 dport,
60 				      int dif, struct hlist_head udptable[])
61 {
62 	struct sock *sk, *result = NULL;
63 	struct hlist_node *node;
64 	unsigned short hnum = ntohs(dport);
65 	int badness = -1;
66 
67 	read_lock(&udp_hash_lock);
68 	sk_for_each(sk, node, &udptable[udp_hashfn(net, hnum)]) {
69 		struct inet_sock *inet = inet_sk(sk);
70 
71 		if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum &&
72 				sk->sk_family == PF_INET6) {
73 			struct ipv6_pinfo *np = inet6_sk(sk);
74 			int score = 0;
75 			if (inet->dport) {
76 				if (inet->dport != sport)
77 					continue;
78 				score++;
79 			}
80 			if (!ipv6_addr_any(&np->rcv_saddr)) {
81 				if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
82 					continue;
83 				score++;
84 			}
85 			if (!ipv6_addr_any(&np->daddr)) {
86 				if (!ipv6_addr_equal(&np->daddr, saddr))
87 					continue;
88 				score++;
89 			}
90 			if (sk->sk_bound_dev_if) {
91 				if (sk->sk_bound_dev_if != dif)
92 					continue;
93 				score++;
94 			}
95 			if (score == 4) {
96 				result = sk;
97 				break;
98 			} else if (score > badness) {
99 				result = sk;
100 				badness = score;
101 			}
102 		}
103 	}
104 	if (result)
105 		sock_hold(result);
106 	read_unlock(&udp_hash_lock);
107 	return result;
108 }
109 
110 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
111 					  __be16 sport, __be16 dport,
112 					  struct hlist_head udptable[])
113 {
114 	struct sock *sk;
115 	struct ipv6hdr *iph = ipv6_hdr(skb);
116 
117 	if (unlikely(sk = skb_steal_sock(skb)))
118 		return sk;
119 	else
120 		return __udp6_lib_lookup(dev_net(skb->dst->dev), &iph->saddr, sport,
121 					 &iph->daddr, dport, inet6_iif(skb),
122 					 udptable);
123 }
124 
125 /*
126  * 	This should be easy, if there is something there we
127  * 	return it, otherwise we block.
128  */
129 
130 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
131 		  struct msghdr *msg, size_t len,
132 		  int noblock, int flags, int *addr_len)
133 {
134 	struct ipv6_pinfo *np = inet6_sk(sk);
135 	struct inet_sock *inet = inet_sk(sk);
136 	struct sk_buff *skb;
137 	unsigned int ulen, copied;
138 	int peeked;
139 	int err;
140 	int is_udplite = IS_UDPLITE(sk);
141 
142 	if (addr_len)
143 		*addr_len=sizeof(struct sockaddr_in6);
144 
145 	if (flags & MSG_ERRQUEUE)
146 		return ipv6_recv_error(sk, msg, len);
147 
148 try_again:
149 	skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
150 				  &peeked, &err);
151 	if (!skb)
152 		goto out;
153 
154 	ulen = skb->len - sizeof(struct udphdr);
155 	copied = len;
156 	if (copied > ulen)
157 		copied = ulen;
158 	else if (copied < ulen)
159 		msg->msg_flags |= MSG_TRUNC;
160 
161 	/*
162 	 * If checksum is needed at all, try to do it while copying the
163 	 * data.  If the data is truncated, or if we only want a partial
164 	 * coverage checksum (UDP-Lite), do it before the copy.
165 	 */
166 
167 	if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
168 		if (udp_lib_checksum_complete(skb))
169 			goto csum_copy_err;
170 	}
171 
172 	if (skb_csum_unnecessary(skb))
173 		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
174 					      msg->msg_iov, copied       );
175 	else {
176 		err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
177 		if (err == -EINVAL)
178 			goto csum_copy_err;
179 	}
180 	if (err)
181 		goto out_free;
182 
183 	if (!peeked)
184 		UDP6_INC_STATS_USER(sock_net(sk),
185 				UDP_MIB_INDATAGRAMS, is_udplite);
186 
187 	sock_recv_timestamp(msg, sk, skb);
188 
189 	/* Copy the address. */
190 	if (msg->msg_name) {
191 		struct sockaddr_in6 *sin6;
192 
193 		sin6 = (struct sockaddr_in6 *) msg->msg_name;
194 		sin6->sin6_family = AF_INET6;
195 		sin6->sin6_port = udp_hdr(skb)->source;
196 		sin6->sin6_flowinfo = 0;
197 		sin6->sin6_scope_id = 0;
198 
199 		if (skb->protocol == htons(ETH_P_IP))
200 			ipv6_addr_set(&sin6->sin6_addr, 0, 0,
201 				      htonl(0xffff), ip_hdr(skb)->saddr);
202 		else {
203 			ipv6_addr_copy(&sin6->sin6_addr,
204 				       &ipv6_hdr(skb)->saddr);
205 			if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
206 				sin6->sin6_scope_id = IP6CB(skb)->iif;
207 		}
208 
209 	}
210 	if (skb->protocol == htons(ETH_P_IP)) {
211 		if (inet->cmsg_flags)
212 			ip_cmsg_recv(msg, skb);
213 	} else {
214 		if (np->rxopt.all)
215 			datagram_recv_ctl(sk, msg, skb);
216 	}
217 
218 	err = copied;
219 	if (flags & MSG_TRUNC)
220 		err = ulen;
221 
222 out_free:
223 	lock_sock(sk);
224 	skb_free_datagram(sk, skb);
225 	release_sock(sk);
226 out:
227 	return err;
228 
229 csum_copy_err:
230 	lock_sock(sk);
231 	if (!skb_kill_datagram(sk, skb, flags))
232 		UDP6_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
233 	release_sock(sk);
234 
235 	if (flags & MSG_DONTWAIT)
236 		return -EAGAIN;
237 	goto try_again;
238 }
239 
240 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
241 		    int type, int code, int offset, __be32 info,
242 		    struct hlist_head udptable[]                    )
243 {
244 	struct ipv6_pinfo *np;
245 	struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
246 	struct in6_addr *saddr = &hdr->saddr;
247 	struct in6_addr *daddr = &hdr->daddr;
248 	struct udphdr *uh = (struct udphdr*)(skb->data+offset);
249 	struct sock *sk;
250 	int err;
251 
252 	sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
253 			       saddr, uh->source, inet6_iif(skb), udptable);
254 	if (sk == NULL)
255 		return;
256 
257 	np = inet6_sk(sk);
258 
259 	if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
260 		goto out;
261 
262 	if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
263 		goto out;
264 
265 	if (np->recverr)
266 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
267 
268 	sk->sk_err = err;
269 	sk->sk_error_report(sk);
270 out:
271 	sock_put(sk);
272 }
273 
274 static __inline__ void udpv6_err(struct sk_buff *skb,
275 				 struct inet6_skb_parm *opt, int type,
276 				 int code, int offset, __be32 info     )
277 {
278 	__udp6_lib_err(skb, opt, type, code, offset, info, udp_hash);
279 }
280 
281 int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
282 {
283 	struct udp_sock *up = udp_sk(sk);
284 	int rc;
285 	int is_udplite = IS_UDPLITE(sk);
286 
287 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
288 		goto drop;
289 
290 	/*
291 	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
292 	 */
293 	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
294 
295 		if (up->pcrlen == 0) {          /* full coverage was set  */
296 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
297 				" %d while full coverage %d requested\n",
298 				UDP_SKB_CB(skb)->cscov, skb->len);
299 			goto drop;
300 		}
301 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
302 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
303 						    "too small, need min %d\n",
304 				       UDP_SKB_CB(skb)->cscov, up->pcrlen);
305 			goto drop;
306 		}
307 	}
308 
309 	if (sk->sk_filter) {
310 		if (udp_lib_checksum_complete(skb))
311 			goto drop;
312 	}
313 
314 	if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
315 		/* Note that an ENOMEM error is charged twice */
316 		if (rc == -ENOMEM) {
317 			UDP6_INC_STATS_BH(sock_net(sk),
318 					UDP_MIB_RCVBUFERRORS, is_udplite);
319 			atomic_inc(&sk->sk_drops);
320 		}
321 		goto drop;
322 	}
323 
324 	return 0;
325 drop:
326 	UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
327 	kfree_skb(skb);
328 	return -1;
329 }
330 
331 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
332 				      __be16 loc_port, struct in6_addr *loc_addr,
333 				      __be16 rmt_port, struct in6_addr *rmt_addr,
334 				      int dif)
335 {
336 	struct hlist_node *node;
337 	struct sock *s = sk;
338 	unsigned short num = ntohs(loc_port);
339 
340 	sk_for_each_from(s, node) {
341 		struct inet_sock *inet = inet_sk(s);
342 
343 		if (!net_eq(sock_net(s), net))
344 			continue;
345 
346 		if (s->sk_hash == num && s->sk_family == PF_INET6) {
347 			struct ipv6_pinfo *np = inet6_sk(s);
348 			if (inet->dport) {
349 				if (inet->dport != rmt_port)
350 					continue;
351 			}
352 			if (!ipv6_addr_any(&np->daddr) &&
353 			    !ipv6_addr_equal(&np->daddr, rmt_addr))
354 				continue;
355 
356 			if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
357 				continue;
358 
359 			if (!ipv6_addr_any(&np->rcv_saddr)) {
360 				if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
361 					continue;
362 			}
363 			if (!inet6_mc_check(s, loc_addr, rmt_addr))
364 				continue;
365 			return s;
366 		}
367 	}
368 	return NULL;
369 }
370 
371 /*
372  * Note: called only from the BH handler context,
373  * so we don't need to lock the hashes.
374  */
375 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
376 		struct in6_addr *saddr, struct in6_addr *daddr,
377 		struct hlist_head udptable[])
378 {
379 	struct sock *sk, *sk2;
380 	const struct udphdr *uh = udp_hdr(skb);
381 	int dif;
382 
383 	read_lock(&udp_hash_lock);
384 	sk = sk_head(&udptable[udp_hashfn(net, ntohs(uh->dest))]);
385 	dif = inet6_iif(skb);
386 	sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
387 	if (!sk) {
388 		kfree_skb(skb);
389 		goto out;
390 	}
391 
392 	sk2 = sk;
393 	while ((sk2 = udp_v6_mcast_next(net, sk_next(sk2), uh->dest, daddr,
394 					uh->source, saddr, dif))) {
395 		struct sk_buff *buff = skb_clone(skb, GFP_ATOMIC);
396 		if (buff) {
397 			bh_lock_sock(sk2);
398 			if (!sock_owned_by_user(sk2))
399 				udpv6_queue_rcv_skb(sk2, buff);
400 			else
401 				sk_add_backlog(sk2, buff);
402 			bh_unlock_sock(sk2);
403 		}
404 	}
405 	bh_lock_sock(sk);
406 	if (!sock_owned_by_user(sk))
407 		udpv6_queue_rcv_skb(sk, skb);
408 	else
409 		sk_add_backlog(sk, skb);
410 	bh_unlock_sock(sk);
411 out:
412 	read_unlock(&udp_hash_lock);
413 	return 0;
414 }
415 
416 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
417 				 int proto)
418 {
419 	int err;
420 
421 	UDP_SKB_CB(skb)->partial_cov = 0;
422 	UDP_SKB_CB(skb)->cscov = skb->len;
423 
424 	if (proto == IPPROTO_UDPLITE) {
425 		err = udplite_checksum_init(skb, uh);
426 		if (err)
427 			return err;
428 	}
429 
430 	if (uh->check == 0) {
431 		/* RFC 2460 section 8.1 says that we SHOULD log
432 		   this error. Well, it is reasonable.
433 		 */
434 		LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
435 		return 1;
436 	}
437 	if (skb->ip_summed == CHECKSUM_COMPLETE &&
438 	    !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
439 			     skb->len, proto, skb->csum))
440 		skb->ip_summed = CHECKSUM_UNNECESSARY;
441 
442 	if (!skb_csum_unnecessary(skb))
443 		skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
444 							 &ipv6_hdr(skb)->daddr,
445 							 skb->len, proto, 0));
446 
447 	return 0;
448 }
449 
450 int __udp6_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
451 		   int proto)
452 {
453 	struct sock *sk;
454 	struct udphdr *uh;
455 	struct net_device *dev = skb->dev;
456 	struct in6_addr *saddr, *daddr;
457 	u32 ulen = 0;
458 	struct net *net = dev_net(skb->dev);
459 
460 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
461 		goto short_packet;
462 
463 	saddr = &ipv6_hdr(skb)->saddr;
464 	daddr = &ipv6_hdr(skb)->daddr;
465 	uh = udp_hdr(skb);
466 
467 	ulen = ntohs(uh->len);
468 	if (ulen > skb->len)
469 		goto short_packet;
470 
471 	if (proto == IPPROTO_UDP) {
472 		/* UDP validates ulen. */
473 
474 		/* Check for jumbo payload */
475 		if (ulen == 0)
476 			ulen = skb->len;
477 
478 		if (ulen < sizeof(*uh))
479 			goto short_packet;
480 
481 		if (ulen < skb->len) {
482 			if (pskb_trim_rcsum(skb, ulen))
483 				goto short_packet;
484 			saddr = &ipv6_hdr(skb)->saddr;
485 			daddr = &ipv6_hdr(skb)->daddr;
486 			uh = udp_hdr(skb);
487 		}
488 	}
489 
490 	if (udp6_csum_init(skb, uh, proto))
491 		goto discard;
492 
493 	/*
494 	 *	Multicast receive code
495 	 */
496 	if (ipv6_addr_is_multicast(daddr))
497 		return __udp6_lib_mcast_deliver(net, skb,
498 				saddr, daddr, udptable);
499 
500 	/* Unicast */
501 
502 	/*
503 	 * check socket cache ... must talk to Alan about his plans
504 	 * for sock caches... i'll skip this for now.
505 	 */
506 	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
507 
508 	if (sk == NULL) {
509 		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
510 			goto discard;
511 
512 		if (udp_lib_checksum_complete(skb))
513 			goto discard;
514 		UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS,
515 				proto == IPPROTO_UDPLITE);
516 
517 		icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0, dev);
518 
519 		kfree_skb(skb);
520 		return 0;
521 	}
522 
523 	/* deliver */
524 
525 	bh_lock_sock(sk);
526 	if (!sock_owned_by_user(sk))
527 		udpv6_queue_rcv_skb(sk, skb);
528 	else
529 		sk_add_backlog(sk, skb);
530 	bh_unlock_sock(sk);
531 	sock_put(sk);
532 	return 0;
533 
534 short_packet:
535 	LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: %d/%u\n",
536 		       proto == IPPROTO_UDPLITE ? "-Lite" : "",
537 		       ulen, skb->len);
538 
539 discard:
540 	UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
541 	kfree_skb(skb);
542 	return 0;
543 }
544 
545 static __inline__ int udpv6_rcv(struct sk_buff *skb)
546 {
547 	return __udp6_lib_rcv(skb, udp_hash, IPPROTO_UDP);
548 }
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 == AF_INET)
558 		udp_flush_pending_frames(sk);
559 	else if (up->pending) {
560 		up->len = 0;
561 		up->pending = 0;
562 		ip6_flush_pending_frames(sk);
563 	}
564 }
565 
566 /*
567  *	Sending
568  */
569 
570 static int udp_v6_push_pending_frames(struct sock *sk)
571 {
572 	struct sk_buff *skb;
573 	struct udphdr *uh;
574 	struct udp_sock  *up = udp_sk(sk);
575 	struct inet_sock *inet = inet_sk(sk);
576 	struct flowi *fl = &inet->cork.fl;
577 	int err = 0;
578 	int is_udplite = IS_UDPLITE(sk);
579 	__wsum csum = 0;
580 
581 	/* Grab the skbuff where UDP header space exists. */
582 	if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
583 		goto out;
584 
585 	/*
586 	 * Create a UDP header
587 	 */
588 	uh = udp_hdr(skb);
589 	uh->source = fl->fl_ip_sport;
590 	uh->dest = fl->fl_ip_dport;
591 	uh->len = htons(up->len);
592 	uh->check = 0;
593 
594 	if (is_udplite)
595 		csum = udplite_csum_outgoing(sk, skb);
596 	 else
597 		csum = udp_csum_outgoing(sk, skb);
598 
599 	/* add protocol-dependent pseudo-header */
600 	uh->check = csum_ipv6_magic(&fl->fl6_src, &fl->fl6_dst,
601 				    up->len, fl->proto, csum   );
602 	if (uh->check == 0)
603 		uh->check = CSUM_MANGLED_0;
604 
605 	err = ip6_push_pending_frames(sk);
606 out:
607 	up->len = 0;
608 	up->pending = 0;
609 	if (!err)
610 		UDP6_INC_STATS_USER(sock_net(sk),
611 				UDP_MIB_OUTDATAGRAMS, is_udplite);
612 	return err;
613 }
614 
615 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
616 		  struct msghdr *msg, size_t len)
617 {
618 	struct ipv6_txoptions opt_space;
619 	struct udp_sock *up = udp_sk(sk);
620 	struct inet_sock *inet = inet_sk(sk);
621 	struct ipv6_pinfo *np = inet6_sk(sk);
622 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
623 	struct in6_addr *daddr, *final_p = NULL, final;
624 	struct ipv6_txoptions *opt = NULL;
625 	struct ip6_flowlabel *flowlabel = NULL;
626 	struct flowi fl;
627 	struct dst_entry *dst;
628 	int addr_len = msg->msg_namelen;
629 	int ulen = len;
630 	int hlimit = -1;
631 	int tclass = -1;
632 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
633 	int err;
634 	int connected = 0;
635 	int is_udplite = IS_UDPLITE(sk);
636 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
637 
638 	/* destination address check */
639 	if (sin6) {
640 		if (addr_len < offsetof(struct sockaddr, sa_data))
641 			return -EINVAL;
642 
643 		switch (sin6->sin6_family) {
644 		case AF_INET6:
645 			if (addr_len < SIN6_LEN_RFC2133)
646 				return -EINVAL;
647 			daddr = &sin6->sin6_addr;
648 			break;
649 		case AF_INET:
650 			goto do_udp_sendmsg;
651 		case AF_UNSPEC:
652 			msg->msg_name = sin6 = NULL;
653 			msg->msg_namelen = addr_len = 0;
654 			daddr = NULL;
655 			break;
656 		default:
657 			return -EINVAL;
658 		}
659 	} else if (!up->pending) {
660 		if (sk->sk_state != TCP_ESTABLISHED)
661 			return -EDESTADDRREQ;
662 		daddr = &np->daddr;
663 	} else
664 		daddr = NULL;
665 
666 	if (daddr) {
667 		if (ipv6_addr_v4mapped(daddr)) {
668 			struct sockaddr_in sin;
669 			sin.sin_family = AF_INET;
670 			sin.sin_port = sin6 ? sin6->sin6_port : inet->dport;
671 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
672 			msg->msg_name = &sin;
673 			msg->msg_namelen = sizeof(sin);
674 do_udp_sendmsg:
675 			if (__ipv6_only_sock(sk))
676 				return -ENETUNREACH;
677 			return udp_sendmsg(iocb, sk, msg, len);
678 		}
679 	}
680 
681 	if (up->pending == AF_INET)
682 		return udp_sendmsg(iocb, sk, msg, len);
683 
684 	/* Rough check on arithmetic overflow,
685 	   better check is made in ip6_append_data().
686 	   */
687 	if (len > INT_MAX - sizeof(struct udphdr))
688 		return -EMSGSIZE;
689 
690 	if (up->pending) {
691 		/*
692 		 * There are pending frames.
693 		 * The socket lock must be held while it's corked.
694 		 */
695 		lock_sock(sk);
696 		if (likely(up->pending)) {
697 			if (unlikely(up->pending != AF_INET6)) {
698 				release_sock(sk);
699 				return -EAFNOSUPPORT;
700 			}
701 			dst = NULL;
702 			goto do_append_data;
703 		}
704 		release_sock(sk);
705 	}
706 	ulen += sizeof(struct udphdr);
707 
708 	memset(&fl, 0, sizeof(fl));
709 
710 	if (sin6) {
711 		if (sin6->sin6_port == 0)
712 			return -EINVAL;
713 
714 		fl.fl_ip_dport = sin6->sin6_port;
715 		daddr = &sin6->sin6_addr;
716 
717 		if (np->sndflow) {
718 			fl.fl6_flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
719 			if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
720 				flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
721 				if (flowlabel == NULL)
722 					return -EINVAL;
723 				daddr = &flowlabel->dst;
724 			}
725 		}
726 
727 		/*
728 		 * Otherwise it will be difficult to maintain
729 		 * sk->sk_dst_cache.
730 		 */
731 		if (sk->sk_state == TCP_ESTABLISHED &&
732 		    ipv6_addr_equal(daddr, &np->daddr))
733 			daddr = &np->daddr;
734 
735 		if (addr_len >= sizeof(struct sockaddr_in6) &&
736 		    sin6->sin6_scope_id &&
737 		    ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
738 			fl.oif = sin6->sin6_scope_id;
739 	} else {
740 		if (sk->sk_state != TCP_ESTABLISHED)
741 			return -EDESTADDRREQ;
742 
743 		fl.fl_ip_dport = inet->dport;
744 		daddr = &np->daddr;
745 		fl.fl6_flowlabel = np->flow_label;
746 		connected = 1;
747 	}
748 
749 	if (!fl.oif)
750 		fl.oif = sk->sk_bound_dev_if;
751 
752 	if (msg->msg_controllen) {
753 		opt = &opt_space;
754 		memset(opt, 0, sizeof(struct ipv6_txoptions));
755 		opt->tot_len = sizeof(*opt);
756 
757 		err = datagram_send_ctl(sock_net(sk), msg, &fl, opt, &hlimit, &tclass);
758 		if (err < 0) {
759 			fl6_sock_release(flowlabel);
760 			return err;
761 		}
762 		if ((fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
763 			flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
764 			if (flowlabel == NULL)
765 				return -EINVAL;
766 		}
767 		if (!(opt->opt_nflen|opt->opt_flen))
768 			opt = NULL;
769 		connected = 0;
770 	}
771 	if (opt == NULL)
772 		opt = np->opt;
773 	if (flowlabel)
774 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
775 	opt = ipv6_fixup_options(&opt_space, opt);
776 
777 	fl.proto = sk->sk_protocol;
778 	if (!ipv6_addr_any(daddr))
779 		ipv6_addr_copy(&fl.fl6_dst, daddr);
780 	else
781 		fl.fl6_dst.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
782 	if (ipv6_addr_any(&fl.fl6_src) && !ipv6_addr_any(&np->saddr))
783 		ipv6_addr_copy(&fl.fl6_src, &np->saddr);
784 	fl.fl_ip_sport = inet->sport;
785 
786 	/* merge ip6_build_xmit from ip6_output */
787 	if (opt && opt->srcrt) {
788 		struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
789 		ipv6_addr_copy(&final, &fl.fl6_dst);
790 		ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
791 		final_p = &final;
792 		connected = 0;
793 	}
794 
795 	if (!fl.oif && ipv6_addr_is_multicast(&fl.fl6_dst)) {
796 		fl.oif = np->mcast_oif;
797 		connected = 0;
798 	}
799 
800 	security_sk_classify_flow(sk, &fl);
801 
802 	err = ip6_sk_dst_lookup(sk, &dst, &fl);
803 	if (err)
804 		goto out;
805 	if (final_p)
806 		ipv6_addr_copy(&fl.fl6_dst, final_p);
807 
808 	if ((err = __xfrm_lookup(&dst, &fl, sk, XFRM_LOOKUP_WAIT)) < 0) {
809 		if (err == -EREMOTE)
810 			err = ip6_dst_blackhole(sk, &dst, &fl);
811 		if (err < 0)
812 			goto out;
813 	}
814 
815 	if (hlimit < 0) {
816 		if (ipv6_addr_is_multicast(&fl.fl6_dst))
817 			hlimit = np->mcast_hops;
818 		else
819 			hlimit = np->hop_limit;
820 		if (hlimit < 0)
821 			hlimit = ip6_dst_hoplimit(dst);
822 	}
823 
824 	if (tclass < 0) {
825 		tclass = np->tclass;
826 		if (tclass < 0)
827 			tclass = 0;
828 	}
829 
830 	if (msg->msg_flags&MSG_CONFIRM)
831 		goto do_confirm;
832 back_from_confirm:
833 
834 	lock_sock(sk);
835 	if (unlikely(up->pending)) {
836 		/* The socket is already corked while preparing it. */
837 		/* ... which is an evident application bug. --ANK */
838 		release_sock(sk);
839 
840 		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
841 		err = -EINVAL;
842 		goto out;
843 	}
844 
845 	up->pending = AF_INET6;
846 
847 do_append_data:
848 	up->len += ulen;
849 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
850 	err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
851 		sizeof(struct udphdr), hlimit, tclass, opt, &fl,
852 		(struct rt6_info*)dst,
853 		corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
854 	if (err)
855 		udp_v6_flush_pending_frames(sk);
856 	else if (!corkreq)
857 		err = udp_v6_push_pending_frames(sk);
858 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
859 		up->pending = 0;
860 
861 	if (dst) {
862 		if (connected) {
863 			ip6_dst_store(sk, dst,
864 				      ipv6_addr_equal(&fl.fl6_dst, &np->daddr) ?
865 				      &np->daddr : NULL,
866 #ifdef CONFIG_IPV6_SUBTREES
867 				      ipv6_addr_equal(&fl.fl6_src, &np->saddr) ?
868 				      &np->saddr :
869 #endif
870 				      NULL);
871 		} else {
872 			dst_release(dst);
873 		}
874 		dst = NULL;
875 	}
876 
877 	if (err > 0)
878 		err = np->recverr ? net_xmit_errno(err) : 0;
879 	release_sock(sk);
880 out:
881 	dst_release(dst);
882 	fl6_sock_release(flowlabel);
883 	if (!err)
884 		return len;
885 	/*
886 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
887 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
888 	 * we don't have a good statistic (IpOutDiscards but it can be too many
889 	 * things).  We could add another new stat but at least for now that
890 	 * seems like overkill.
891 	 */
892 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
893 		UDP6_INC_STATS_USER(sock_net(sk),
894 				UDP_MIB_SNDBUFERRORS, is_udplite);
895 	}
896 	return err;
897 
898 do_confirm:
899 	dst_confirm(dst);
900 	if (!(msg->msg_flags&MSG_PROBE) || len)
901 		goto back_from_confirm;
902 	err = 0;
903 	goto out;
904 }
905 
906 void udpv6_destroy_sock(struct sock *sk)
907 {
908 	lock_sock(sk);
909 	udp_v6_flush_pending_frames(sk);
910 	release_sock(sk);
911 
912 	inet6_destroy_sock(sk);
913 }
914 
915 /*
916  *	Socket option code for UDP
917  */
918 int udpv6_setsockopt(struct sock *sk, int level, int optname,
919 		     char __user *optval, int optlen)
920 {
921 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
922 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
923 					  udp_v6_push_pending_frames);
924 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
925 }
926 
927 #ifdef CONFIG_COMPAT
928 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
929 			    char __user *optval, int optlen)
930 {
931 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
932 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
933 					  udp_v6_push_pending_frames);
934 	return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
935 }
936 #endif
937 
938 int udpv6_getsockopt(struct sock *sk, int level, int optname,
939 		     char __user *optval, int __user *optlen)
940 {
941 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
942 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
943 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
944 }
945 
946 #ifdef CONFIG_COMPAT
947 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
948 			    char __user *optval, int __user *optlen)
949 {
950 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
951 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
952 	return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
953 }
954 #endif
955 
956 static struct inet6_protocol udpv6_protocol = {
957 	.handler	=	udpv6_rcv,
958 	.err_handler	=	udpv6_err,
959 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
960 };
961 
962 /* ------------------------------------------------------------------------ */
963 #ifdef CONFIG_PROC_FS
964 
965 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
966 {
967 	struct inet_sock *inet = inet_sk(sp);
968 	struct ipv6_pinfo *np = inet6_sk(sp);
969 	struct in6_addr *dest, *src;
970 	__u16 destp, srcp;
971 
972 	dest  = &np->daddr;
973 	src   = &np->rcv_saddr;
974 	destp = ntohs(inet->dport);
975 	srcp  = ntohs(inet->sport);
976 	seq_printf(seq,
977 		   "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
978 		   "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d\n",
979 		   bucket,
980 		   src->s6_addr32[0], src->s6_addr32[1],
981 		   src->s6_addr32[2], src->s6_addr32[3], srcp,
982 		   dest->s6_addr32[0], dest->s6_addr32[1],
983 		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
984 		   sp->sk_state,
985 		   atomic_read(&sp->sk_wmem_alloc),
986 		   atomic_read(&sp->sk_rmem_alloc),
987 		   0, 0L, 0,
988 		   sock_i_uid(sp), 0,
989 		   sock_i_ino(sp),
990 		   atomic_read(&sp->sk_refcnt), sp,
991 		   atomic_read(&sp->sk_drops));
992 }
993 
994 int udp6_seq_show(struct seq_file *seq, void *v)
995 {
996 	if (v == SEQ_START_TOKEN)
997 		seq_printf(seq,
998 			   "  sl  "
999 			   "local_address                         "
1000 			   "remote_address                        "
1001 			   "st tx_queue rx_queue tr tm->when retrnsmt"
1002 			   "   uid  timeout inode ref pointer drops\n");
1003 	else
1004 		udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1005 	return 0;
1006 }
1007 
1008 static struct udp_seq_afinfo udp6_seq_afinfo = {
1009 	.name		= "udp6",
1010 	.family		= AF_INET6,
1011 	.hashtable	= udp_hash,
1012 	.seq_fops	= {
1013 		.owner	=	THIS_MODULE,
1014 	},
1015 	.seq_ops	= {
1016 		.show		= udp6_seq_show,
1017 	},
1018 };
1019 
1020 int udp6_proc_init(struct net *net)
1021 {
1022 	return udp_proc_register(net, &udp6_seq_afinfo);
1023 }
1024 
1025 void udp6_proc_exit(struct net *net) {
1026 	udp_proc_unregister(net, &udp6_seq_afinfo);
1027 }
1028 #endif /* CONFIG_PROC_FS */
1029 
1030 /* ------------------------------------------------------------------------ */
1031 
1032 struct proto udpv6_prot = {
1033 	.name		   = "UDPv6",
1034 	.owner		   = THIS_MODULE,
1035 	.close		   = udp_lib_close,
1036 	.connect	   = ip6_datagram_connect,
1037 	.disconnect	   = udp_disconnect,
1038 	.ioctl		   = udp_ioctl,
1039 	.destroy	   = udpv6_destroy_sock,
1040 	.setsockopt	   = udpv6_setsockopt,
1041 	.getsockopt	   = udpv6_getsockopt,
1042 	.sendmsg	   = udpv6_sendmsg,
1043 	.recvmsg	   = udpv6_recvmsg,
1044 	.backlog_rcv	   = udpv6_queue_rcv_skb,
1045 	.hash		   = udp_lib_hash,
1046 	.unhash		   = udp_lib_unhash,
1047 	.get_port	   = udp_v6_get_port,
1048 	.memory_allocated  = &udp_memory_allocated,
1049 	.sysctl_mem	   = sysctl_udp_mem,
1050 	.sysctl_wmem	   = &sysctl_udp_wmem_min,
1051 	.sysctl_rmem	   = &sysctl_udp_rmem_min,
1052 	.obj_size	   = sizeof(struct udp6_sock),
1053 	.h.udp_hash	   = udp_hash,
1054 #ifdef CONFIG_COMPAT
1055 	.compat_setsockopt = compat_udpv6_setsockopt,
1056 	.compat_getsockopt = compat_udpv6_getsockopt,
1057 #endif
1058 };
1059 
1060 static struct inet_protosw udpv6_protosw = {
1061 	.type =      SOCK_DGRAM,
1062 	.protocol =  IPPROTO_UDP,
1063 	.prot =      &udpv6_prot,
1064 	.ops =       &inet6_dgram_ops,
1065 	.capability =-1,
1066 	.no_check =  UDP_CSUM_DEFAULT,
1067 	.flags =     INET_PROTOSW_PERMANENT,
1068 };
1069 
1070 
1071 int __init udpv6_init(void)
1072 {
1073 	int ret;
1074 
1075 	ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1076 	if (ret)
1077 		goto out;
1078 
1079 	ret = inet6_register_protosw(&udpv6_protosw);
1080 	if (ret)
1081 		goto out_udpv6_protocol;
1082 out:
1083 	return ret;
1084 
1085 out_udpv6_protocol:
1086 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1087 	goto out;
1088 }
1089 
1090 void udpv6_exit(void)
1091 {
1092 	inet6_unregister_protosw(&udpv6_protosw);
1093 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1094 }
1095