xref: /linux/net/ipv6/udp.c (revision c537b994505099b7197e7d3125b942ecbcc51eb6)
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/errno.h>
27 #include <linux/types.h>
28 #include <linux/socket.h>
29 #include <linux/sockios.h>
30 #include <linux/net.h>
31 #include <linux/in6.h>
32 #include <linux/netdevice.h>
33 #include <linux/if_arp.h>
34 #include <linux/ipv6.h>
35 #include <linux/icmpv6.h>
36 #include <linux/init.h>
37 #include <linux/skbuff.h>
38 #include <asm/uaccess.h>
39 
40 #include <net/ndisc.h>
41 #include <net/protocol.h>
42 #include <net/transp_v6.h>
43 #include <net/ip6_route.h>
44 #include <net/raw.h>
45 #include <net/tcp_states.h>
46 #include <net/ip6_checksum.h>
47 #include <net/xfrm.h>
48 
49 #include <linux/proc_fs.h>
50 #include <linux/seq_file.h>
51 #include "udp_impl.h"
52 
53 DEFINE_SNMP_STAT(struct udp_mib, udp_stats_in6) __read_mostly;
54 
55 static inline int udp_v6_get_port(struct sock *sk, unsigned short snum)
56 {
57 	return udp_get_port(sk, snum, ipv6_rcv_saddr_equal);
58 }
59 
60 static struct sock *__udp6_lib_lookup(struct in6_addr *saddr, __be16 sport,
61 				      struct in6_addr *daddr, __be16 dport,
62 				      int dif, struct hlist_head udptable[])
63 {
64 	struct sock *sk, *result = NULL;
65 	struct hlist_node *node;
66 	unsigned short hnum = ntohs(dport);
67 	int badness = -1;
68 
69 	read_lock(&udp_hash_lock);
70 	sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
71 		struct inet_sock *inet = inet_sk(sk);
72 
73 		if (sk->sk_hash == hnum && sk->sk_family == PF_INET6) {
74 			struct ipv6_pinfo *np = inet6_sk(sk);
75 			int score = 0;
76 			if (inet->dport) {
77 				if (inet->dport != sport)
78 					continue;
79 				score++;
80 			}
81 			if (!ipv6_addr_any(&np->rcv_saddr)) {
82 				if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
83 					continue;
84 				score++;
85 			}
86 			if (!ipv6_addr_any(&np->daddr)) {
87 				if (!ipv6_addr_equal(&np->daddr, saddr))
88 					continue;
89 				score++;
90 			}
91 			if (sk->sk_bound_dev_if) {
92 				if (sk->sk_bound_dev_if != dif)
93 					continue;
94 				score++;
95 			}
96 			if(score == 4) {
97 				result = sk;
98 				break;
99 			} else if(score > badness) {
100 				result = sk;
101 				badness = score;
102 			}
103 		}
104 	}
105 	if (result)
106 		sock_hold(result);
107 	read_unlock(&udp_hash_lock);
108 	return result;
109 }
110 
111 /*
112  * 	This should be easy, if there is something there we
113  * 	return it, otherwise we block.
114  */
115 
116 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
117 		  struct msghdr *msg, size_t len,
118 		  int noblock, int flags, int *addr_len)
119 {
120 	struct ipv6_pinfo *np = inet6_sk(sk);
121 	struct inet_sock *inet = inet_sk(sk);
122 	struct sk_buff *skb;
123 	size_t copied;
124 	int err, copy_only, is_udplite = IS_UDPLITE(sk);
125 
126 	if (addr_len)
127 		*addr_len=sizeof(struct sockaddr_in6);
128 
129 	if (flags & MSG_ERRQUEUE)
130 		return ipv6_recv_error(sk, msg, len);
131 
132 try_again:
133 	skb = skb_recv_datagram(sk, flags, noblock, &err);
134 	if (!skb)
135 		goto out;
136 
137 	copied = skb->len - sizeof(struct udphdr);
138 	if (copied > len) {
139 		copied = len;
140 		msg->msg_flags |= MSG_TRUNC;
141 	}
142 
143 	/*
144 	 * 	Decide whether to checksum and/or copy data.
145 	 */
146 	copy_only = (skb->ip_summed==CHECKSUM_UNNECESSARY);
147 
148 	if (is_udplite  ||  (!copy_only  &&  msg->msg_flags&MSG_TRUNC)) {
149 		if (__udp_lib_checksum_complete(skb))
150 			goto csum_copy_err;
151 		copy_only = 1;
152 	}
153 
154 	if (copy_only)
155 		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
156 					      msg->msg_iov, copied       );
157 	else {
158 		err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
159 		if (err == -EINVAL)
160 			goto csum_copy_err;
161 	}
162 	if (err)
163 		goto out_free;
164 
165 	sock_recv_timestamp(msg, sk, skb);
166 
167 	/* Copy the address. */
168 	if (msg->msg_name) {
169 		struct sockaddr_in6 *sin6;
170 
171 		sin6 = (struct sockaddr_in6 *) msg->msg_name;
172 		sin6->sin6_family = AF_INET6;
173 		sin6->sin6_port = skb->h.uh->source;
174 		sin6->sin6_flowinfo = 0;
175 		sin6->sin6_scope_id = 0;
176 
177 		if (skb->protocol == htons(ETH_P_IP))
178 			ipv6_addr_set(&sin6->sin6_addr, 0, 0,
179 				      htonl(0xffff), skb->nh.iph->saddr);
180 		else {
181 			ipv6_addr_copy(&sin6->sin6_addr, &skb->nh.ipv6h->saddr);
182 			if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
183 				sin6->sin6_scope_id = IP6CB(skb)->iif;
184 		}
185 
186 	}
187 	if (skb->protocol == htons(ETH_P_IP)) {
188 		if (inet->cmsg_flags)
189 			ip_cmsg_recv(msg, skb);
190 	} else {
191 		if (np->rxopt.all)
192 			datagram_recv_ctl(sk, msg, skb);
193 	}
194 
195 	err = copied;
196 	if (flags & MSG_TRUNC)
197 		err = skb->len - sizeof(struct udphdr);
198 
199 out_free:
200 	skb_free_datagram(sk, skb);
201 out:
202 	return err;
203 
204 csum_copy_err:
205 	skb_kill_datagram(sk, skb, flags);
206 
207 	if (flags & MSG_DONTWAIT) {
208 		UDP6_INC_STATS_USER(UDP_MIB_INERRORS, is_udplite);
209 		return -EAGAIN;
210 	}
211 	goto try_again;
212 }
213 
214 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
215 		    int type, int code, int offset, __be32 info,
216 		    struct hlist_head udptable[]                    )
217 {
218 	struct ipv6_pinfo *np;
219 	struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
220 	struct in6_addr *saddr = &hdr->saddr;
221 	struct in6_addr *daddr = &hdr->daddr;
222 	struct udphdr *uh = (struct udphdr*)(skb->data+offset);
223 	struct sock *sk;
224 	int err;
225 
226 	sk = __udp6_lib_lookup(daddr, uh->dest,
227 			       saddr, uh->source, inet6_iif(skb), udptable);
228 	if (sk == NULL)
229 		return;
230 
231 	np = inet6_sk(sk);
232 
233 	if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
234 		goto out;
235 
236 	if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
237 		goto out;
238 
239 	if (np->recverr)
240 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
241 
242 	sk->sk_err = err;
243 	sk->sk_error_report(sk);
244 out:
245 	sock_put(sk);
246 }
247 
248 static __inline__ void udpv6_err(struct sk_buff *skb,
249 				 struct inet6_skb_parm *opt, int type,
250 				 int code, int offset, __be32 info     )
251 {
252 	return __udp6_lib_err(skb, opt, type, code, offset, info, udp_hash);
253 }
254 
255 int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
256 {
257 	struct udp_sock *up = udp_sk(sk);
258 	int rc;
259 
260 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
261 		goto drop;
262 
263 	/*
264 	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
265 	 */
266 	if ((up->pcflag & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
267 
268 		if (up->pcrlen == 0) {          /* full coverage was set  */
269 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
270 				" %d while full coverage %d requested\n",
271 				UDP_SKB_CB(skb)->cscov, skb->len);
272 			goto drop;
273 		}
274 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
275 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
276 						    "too small, need min %d\n",
277 				       UDP_SKB_CB(skb)->cscov, up->pcrlen);
278 			goto drop;
279 		}
280 	}
281 
282 	if (udp_lib_checksum_complete(skb))
283 		goto drop;
284 
285 	if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
286 		/* Note that an ENOMEM error is charged twice */
287 		if (rc == -ENOMEM)
288 			UDP6_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, up->pcflag);
289 		goto drop;
290 	}
291 	UDP6_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
292 	return 0;
293 drop:
294 	UDP6_INC_STATS_BH(UDP_MIB_INERRORS, up->pcflag);
295 	kfree_skb(skb);
296 	return -1;
297 }
298 
299 static struct sock *udp_v6_mcast_next(struct sock *sk,
300 				      __be16 loc_port, struct in6_addr *loc_addr,
301 				      __be16 rmt_port, struct in6_addr *rmt_addr,
302 				      int dif)
303 {
304 	struct hlist_node *node;
305 	struct sock *s = sk;
306 	unsigned short num = ntohs(loc_port);
307 
308 	sk_for_each_from(s, node) {
309 		struct inet_sock *inet = inet_sk(s);
310 
311 		if (s->sk_hash == num && s->sk_family == PF_INET6) {
312 			struct ipv6_pinfo *np = inet6_sk(s);
313 			if (inet->dport) {
314 				if (inet->dport != rmt_port)
315 					continue;
316 			}
317 			if (!ipv6_addr_any(&np->daddr) &&
318 			    !ipv6_addr_equal(&np->daddr, rmt_addr))
319 				continue;
320 
321 			if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
322 				continue;
323 
324 			if (!ipv6_addr_any(&np->rcv_saddr)) {
325 				if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
326 					continue;
327 			}
328 			if(!inet6_mc_check(s, loc_addr, rmt_addr))
329 				continue;
330 			return s;
331 		}
332 	}
333 	return NULL;
334 }
335 
336 /*
337  * Note: called only from the BH handler context,
338  * so we don't need to lock the hashes.
339  */
340 static int __udp6_lib_mcast_deliver(struct sk_buff *skb, struct in6_addr *saddr,
341 			   struct in6_addr *daddr, struct hlist_head udptable[])
342 {
343 	struct sock *sk, *sk2;
344 	const struct udphdr *uh = skb->h.uh;
345 	int dif;
346 
347 	read_lock(&udp_hash_lock);
348 	sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
349 	dif = inet6_iif(skb);
350 	sk = udp_v6_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
351 	if (!sk) {
352 		kfree_skb(skb);
353 		goto out;
354 	}
355 
356 	sk2 = sk;
357 	while ((sk2 = udp_v6_mcast_next(sk_next(sk2), uh->dest, daddr,
358 					uh->source, saddr, dif))) {
359 		struct sk_buff *buff = skb_clone(skb, GFP_ATOMIC);
360 		if (buff)
361 			udpv6_queue_rcv_skb(sk2, buff);
362 	}
363 	udpv6_queue_rcv_skb(sk, skb);
364 out:
365 	read_unlock(&udp_hash_lock);
366 	return 0;
367 }
368 
369 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh)
370 
371 {
372 	if (uh->check == 0) {
373 		/* RFC 2460 section 8.1 says that we SHOULD log
374 		   this error. Well, it is reasonable.
375 		 */
376 		LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
377 		return 1;
378 	}
379 	if (skb->ip_summed == CHECKSUM_COMPLETE &&
380 	    !csum_ipv6_magic(&skb->nh.ipv6h->saddr, &skb->nh.ipv6h->daddr,
381 			     skb->len, IPPROTO_UDP, skb->csum             ))
382 		skb->ip_summed = CHECKSUM_UNNECESSARY;
383 
384 	if (skb->ip_summed != CHECKSUM_UNNECESSARY)
385 		skb->csum = ~csum_unfold(csum_ipv6_magic(&skb->nh.ipv6h->saddr,
386 							 &skb->nh.ipv6h->daddr,
387 							 skb->len, IPPROTO_UDP,
388 							 0));
389 
390 	return (UDP_SKB_CB(skb)->partial_cov = 0);
391 }
392 
393 int __udp6_lib_rcv(struct sk_buff **pskb, struct hlist_head udptable[],
394 		   int is_udplite)
395 {
396 	struct sk_buff *skb = *pskb;
397 	struct sock *sk;
398 	struct udphdr *uh;
399 	struct net_device *dev = skb->dev;
400 	struct in6_addr *saddr, *daddr;
401 	u32 ulen = 0;
402 
403 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
404 		goto short_packet;
405 
406 	saddr = &skb->nh.ipv6h->saddr;
407 	daddr = &skb->nh.ipv6h->daddr;
408 	uh = skb->h.uh;
409 
410 	ulen = ntohs(uh->len);
411 	if (ulen > skb->len)
412 		goto short_packet;
413 
414 	if(! is_udplite ) {		/* UDP validates ulen. */
415 
416 		/* Check for jumbo payload */
417 		if (ulen == 0)
418 			ulen = skb->len;
419 
420 		if (ulen < sizeof(*uh))
421 			goto short_packet;
422 
423 		if (ulen < skb->len) {
424 			if (pskb_trim_rcsum(skb, ulen))
425 				goto short_packet;
426 			saddr = &skb->nh.ipv6h->saddr;
427 			daddr = &skb->nh.ipv6h->daddr;
428 			uh = skb->h.uh;
429 		}
430 
431 		if (udp6_csum_init(skb, uh))
432 			goto discard;
433 
434 	} else 	{			/* UDP-Lite validates cscov. */
435 		if (udplite6_csum_init(skb, uh))
436 			goto discard;
437 	}
438 
439 	/*
440 	 *	Multicast receive code
441 	 */
442 	if (ipv6_addr_is_multicast(daddr))
443 		return __udp6_lib_mcast_deliver(skb, saddr, daddr, udptable);
444 
445 	/* Unicast */
446 
447 	/*
448 	 * check socket cache ... must talk to Alan about his plans
449 	 * for sock caches... i'll skip this for now.
450 	 */
451 	sk = __udp6_lib_lookup(saddr, uh->source,
452 			       daddr, uh->dest, inet6_iif(skb), udptable);
453 
454 	if (sk == NULL) {
455 		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
456 			goto discard;
457 
458 		if (udp_lib_checksum_complete(skb))
459 			goto discard;
460 		UDP6_INC_STATS_BH(UDP_MIB_NOPORTS, is_udplite);
461 
462 		icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0, dev);
463 
464 		kfree_skb(skb);
465 		return(0);
466 	}
467 
468 	/* deliver */
469 
470 	udpv6_queue_rcv_skb(sk, skb);
471 	sock_put(sk);
472 	return(0);
473 
474 short_packet:
475 	LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: %d/%u\n",
476 		       is_udplite? "-Lite" : "",  ulen, skb->len);
477 
478 discard:
479 	UDP6_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
480 	kfree_skb(skb);
481 	return(0);
482 }
483 
484 static __inline__ int udpv6_rcv(struct sk_buff **pskb)
485 {
486 	return __udp6_lib_rcv(pskb, udp_hash, 0);
487 }
488 
489 /*
490  * Throw away all pending data and cancel the corking. Socket is locked.
491  */
492 static void udp_v6_flush_pending_frames(struct sock *sk)
493 {
494 	struct udp_sock *up = udp_sk(sk);
495 
496 	if (up->pending) {
497 		up->len = 0;
498 		up->pending = 0;
499 		ip6_flush_pending_frames(sk);
500 	}
501 }
502 
503 /*
504  *	Sending
505  */
506 
507 static int udp_v6_push_pending_frames(struct sock *sk)
508 {
509 	struct sk_buff *skb;
510 	struct udphdr *uh;
511 	struct udp_sock  *up = udp_sk(sk);
512 	struct inet_sock *inet = inet_sk(sk);
513 	struct flowi *fl = &inet->cork.fl;
514 	int err = 0;
515 	__wsum csum = 0;
516 
517 	/* Grab the skbuff where UDP header space exists. */
518 	if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
519 		goto out;
520 
521 	/*
522 	 * Create a UDP header
523 	 */
524 	uh = skb->h.uh;
525 	uh->source = fl->fl_ip_sport;
526 	uh->dest = fl->fl_ip_dport;
527 	uh->len = htons(up->len);
528 	uh->check = 0;
529 
530 	if (up->pcflag)
531 		csum = udplite_csum_outgoing(sk, skb);
532 	 else
533 		csum = udp_csum_outgoing(sk, skb);
534 
535 	/* add protocol-dependent pseudo-header */
536 	uh->check = csum_ipv6_magic(&fl->fl6_src, &fl->fl6_dst,
537 				    up->len, fl->proto, csum   );
538 	if (uh->check == 0)
539 		uh->check = CSUM_MANGLED_0;
540 
541 	err = ip6_push_pending_frames(sk);
542 out:
543 	up->len = 0;
544 	up->pending = 0;
545 	return err;
546 }
547 
548 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
549 		  struct msghdr *msg, size_t len)
550 {
551 	struct ipv6_txoptions opt_space;
552 	struct udp_sock *up = udp_sk(sk);
553 	struct inet_sock *inet = inet_sk(sk);
554 	struct ipv6_pinfo *np = inet6_sk(sk);
555 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
556 	struct in6_addr *daddr, *final_p = NULL, final;
557 	struct ipv6_txoptions *opt = NULL;
558 	struct ip6_flowlabel *flowlabel = NULL;
559 	struct flowi fl;
560 	struct dst_entry *dst;
561 	int addr_len = msg->msg_namelen;
562 	int ulen = len;
563 	int hlimit = -1;
564 	int tclass = -1;
565 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
566 	int err;
567 	int connected = 0;
568 	int is_udplite = up->pcflag;
569 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
570 
571 	/* destination address check */
572 	if (sin6) {
573 		if (addr_len < offsetof(struct sockaddr, sa_data))
574 			return -EINVAL;
575 
576 		switch (sin6->sin6_family) {
577 		case AF_INET6:
578 			if (addr_len < SIN6_LEN_RFC2133)
579 				return -EINVAL;
580 			daddr = &sin6->sin6_addr;
581 			break;
582 		case AF_INET:
583 			goto do_udp_sendmsg;
584 		case AF_UNSPEC:
585 			msg->msg_name = sin6 = NULL;
586 			msg->msg_namelen = addr_len = 0;
587 			daddr = NULL;
588 			break;
589 		default:
590 			return -EINVAL;
591 		}
592 	} else if (!up->pending) {
593 		if (sk->sk_state != TCP_ESTABLISHED)
594 			return -EDESTADDRREQ;
595 		daddr = &np->daddr;
596 	} else
597 		daddr = NULL;
598 
599 	if (daddr) {
600 		if (ipv6_addr_type(daddr) == IPV6_ADDR_MAPPED) {
601 			struct sockaddr_in sin;
602 			sin.sin_family = AF_INET;
603 			sin.sin_port = sin6 ? sin6->sin6_port : inet->dport;
604 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
605 			msg->msg_name = &sin;
606 			msg->msg_namelen = sizeof(sin);
607 do_udp_sendmsg:
608 			if (__ipv6_only_sock(sk))
609 				return -ENETUNREACH;
610 			return udp_sendmsg(iocb, sk, msg, len);
611 		}
612 	}
613 
614 	if (up->pending == AF_INET)
615 		return udp_sendmsg(iocb, sk, msg, len);
616 
617 	/* Rough check on arithmetic overflow,
618 	   better check is made in ip6_build_xmit
619 	   */
620 	if (len > INT_MAX - sizeof(struct udphdr))
621 		return -EMSGSIZE;
622 
623 	if (up->pending) {
624 		/*
625 		 * There are pending frames.
626 		 * The socket lock must be held while it's corked.
627 		 */
628 		lock_sock(sk);
629 		if (likely(up->pending)) {
630 			if (unlikely(up->pending != AF_INET6)) {
631 				release_sock(sk);
632 				return -EAFNOSUPPORT;
633 			}
634 			dst = NULL;
635 			goto do_append_data;
636 		}
637 		release_sock(sk);
638 	}
639 	ulen += sizeof(struct udphdr);
640 
641 	memset(&fl, 0, sizeof(fl));
642 
643 	if (sin6) {
644 		if (sin6->sin6_port == 0)
645 			return -EINVAL;
646 
647 		fl.fl_ip_dport = sin6->sin6_port;
648 		daddr = &sin6->sin6_addr;
649 
650 		if (np->sndflow) {
651 			fl.fl6_flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
652 			if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
653 				flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
654 				if (flowlabel == NULL)
655 					return -EINVAL;
656 				daddr = &flowlabel->dst;
657 			}
658 		}
659 
660 		/*
661 		 * Otherwise it will be difficult to maintain
662 		 * sk->sk_dst_cache.
663 		 */
664 		if (sk->sk_state == TCP_ESTABLISHED &&
665 		    ipv6_addr_equal(daddr, &np->daddr))
666 			daddr = &np->daddr;
667 
668 		if (addr_len >= sizeof(struct sockaddr_in6) &&
669 		    sin6->sin6_scope_id &&
670 		    ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
671 			fl.oif = sin6->sin6_scope_id;
672 	} else {
673 		if (sk->sk_state != TCP_ESTABLISHED)
674 			return -EDESTADDRREQ;
675 
676 		fl.fl_ip_dport = inet->dport;
677 		daddr = &np->daddr;
678 		fl.fl6_flowlabel = np->flow_label;
679 		connected = 1;
680 	}
681 
682 	if (!fl.oif)
683 		fl.oif = sk->sk_bound_dev_if;
684 
685 	if (msg->msg_controllen) {
686 		opt = &opt_space;
687 		memset(opt, 0, sizeof(struct ipv6_txoptions));
688 		opt->tot_len = sizeof(*opt);
689 
690 		err = datagram_send_ctl(msg, &fl, opt, &hlimit, &tclass);
691 		if (err < 0) {
692 			fl6_sock_release(flowlabel);
693 			return err;
694 		}
695 		if ((fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
696 			flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
697 			if (flowlabel == NULL)
698 				return -EINVAL;
699 		}
700 		if (!(opt->opt_nflen|opt->opt_flen))
701 			opt = NULL;
702 		connected = 0;
703 	}
704 	if (opt == NULL)
705 		opt = np->opt;
706 	if (flowlabel)
707 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
708 	opt = ipv6_fixup_options(&opt_space, opt);
709 
710 	fl.proto = sk->sk_protocol;
711 	ipv6_addr_copy(&fl.fl6_dst, daddr);
712 	if (ipv6_addr_any(&fl.fl6_src) && !ipv6_addr_any(&np->saddr))
713 		ipv6_addr_copy(&fl.fl6_src, &np->saddr);
714 	fl.fl_ip_sport = inet->sport;
715 
716 	/* merge ip6_build_xmit from ip6_output */
717 	if (opt && opt->srcrt) {
718 		struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
719 		ipv6_addr_copy(&final, &fl.fl6_dst);
720 		ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
721 		final_p = &final;
722 		connected = 0;
723 	}
724 
725 	if (!fl.oif && ipv6_addr_is_multicast(&fl.fl6_dst)) {
726 		fl.oif = np->mcast_oif;
727 		connected = 0;
728 	}
729 
730 	security_sk_classify_flow(sk, &fl);
731 
732 	err = ip6_sk_dst_lookup(sk, &dst, &fl);
733 	if (err)
734 		goto out;
735 	if (final_p)
736 		ipv6_addr_copy(&fl.fl6_dst, final_p);
737 
738 	if ((err = xfrm_lookup(&dst, &fl, sk, 1)) < 0)
739 		goto out;
740 
741 	if (hlimit < 0) {
742 		if (ipv6_addr_is_multicast(&fl.fl6_dst))
743 			hlimit = np->mcast_hops;
744 		else
745 			hlimit = np->hop_limit;
746 		if (hlimit < 0)
747 			hlimit = dst_metric(dst, RTAX_HOPLIMIT);
748 		if (hlimit < 0)
749 			hlimit = ipv6_get_hoplimit(dst->dev);
750 	}
751 
752 	if (tclass < 0) {
753 		tclass = np->tclass;
754 		if (tclass < 0)
755 			tclass = 0;
756 	}
757 
758 	if (msg->msg_flags&MSG_CONFIRM)
759 		goto do_confirm;
760 back_from_confirm:
761 
762 	lock_sock(sk);
763 	if (unlikely(up->pending)) {
764 		/* The socket is already corked while preparing it. */
765 		/* ... which is an evident application bug. --ANK */
766 		release_sock(sk);
767 
768 		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
769 		err = -EINVAL;
770 		goto out;
771 	}
772 
773 	up->pending = AF_INET6;
774 
775 do_append_data:
776 	up->len += ulen;
777 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
778 	err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
779 		sizeof(struct udphdr), hlimit, tclass, opt, &fl,
780 		(struct rt6_info*)dst,
781 		corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
782 	if (err)
783 		udp_v6_flush_pending_frames(sk);
784 	else if (!corkreq)
785 		err = udp_v6_push_pending_frames(sk);
786 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
787 		up->pending = 0;
788 
789 	if (dst) {
790 		if (connected) {
791 			ip6_dst_store(sk, dst,
792 				      ipv6_addr_equal(&fl.fl6_dst, &np->daddr) ?
793 				      &np->daddr : NULL,
794 #ifdef CONFIG_IPV6_SUBTREES
795 				      ipv6_addr_equal(&fl.fl6_src, &np->saddr) ?
796 				      &np->saddr :
797 #endif
798 				      NULL);
799 		} else {
800 			dst_release(dst);
801 		}
802 	}
803 
804 	if (err > 0)
805 		err = np->recverr ? net_xmit_errno(err) : 0;
806 	release_sock(sk);
807 out:
808 	fl6_sock_release(flowlabel);
809 	if (!err) {
810 		UDP6_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
811 		return len;
812 	}
813 	/*
814 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
815 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
816 	 * we don't have a good statistic (IpOutDiscards but it can be too many
817 	 * things).  We could add another new stat but at least for now that
818 	 * seems like overkill.
819 	 */
820 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
821 		UDP6_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
822 	}
823 	return err;
824 
825 do_confirm:
826 	dst_confirm(dst);
827 	if (!(msg->msg_flags&MSG_PROBE) || len)
828 		goto back_from_confirm;
829 	err = 0;
830 	goto out;
831 }
832 
833 int udpv6_destroy_sock(struct sock *sk)
834 {
835 	lock_sock(sk);
836 	udp_v6_flush_pending_frames(sk);
837 	release_sock(sk);
838 
839 	inet6_destroy_sock(sk);
840 
841 	return 0;
842 }
843 
844 /*
845  *	Socket option code for UDP
846  */
847 int udpv6_setsockopt(struct sock *sk, int level, int optname,
848 		     char __user *optval, int optlen)
849 {
850 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
851 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
852 					  udp_v6_push_pending_frames);
853 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
854 }
855 
856 #ifdef CONFIG_COMPAT
857 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
858 			    char __user *optval, int optlen)
859 {
860 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
861 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
862 					  udp_v6_push_pending_frames);
863 	return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
864 }
865 #endif
866 
867 int udpv6_getsockopt(struct sock *sk, int level, int optname,
868 		     char __user *optval, int __user *optlen)
869 {
870 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
871 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
872 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
873 }
874 
875 #ifdef CONFIG_COMPAT
876 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
877 			    char __user *optval, int __user *optlen)
878 {
879 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
880 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
881 	return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
882 }
883 #endif
884 
885 static struct inet6_protocol udpv6_protocol = {
886 	.handler	=	udpv6_rcv,
887 	.err_handler	=	udpv6_err,
888 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
889 };
890 
891 /* ------------------------------------------------------------------------ */
892 #ifdef CONFIG_PROC_FS
893 
894 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
895 {
896 	struct inet_sock *inet = inet_sk(sp);
897 	struct ipv6_pinfo *np = inet6_sk(sp);
898 	struct in6_addr *dest, *src;
899 	__u16 destp, srcp;
900 
901 	dest  = &np->daddr;
902 	src   = &np->rcv_saddr;
903 	destp = ntohs(inet->dport);
904 	srcp  = ntohs(inet->sport);
905 	seq_printf(seq,
906 		   "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
907 		   "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p\n",
908 		   bucket,
909 		   src->s6_addr32[0], src->s6_addr32[1],
910 		   src->s6_addr32[2], src->s6_addr32[3], srcp,
911 		   dest->s6_addr32[0], dest->s6_addr32[1],
912 		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
913 		   sp->sk_state,
914 		   atomic_read(&sp->sk_wmem_alloc),
915 		   atomic_read(&sp->sk_rmem_alloc),
916 		   0, 0L, 0,
917 		   sock_i_uid(sp), 0,
918 		   sock_i_ino(sp),
919 		   atomic_read(&sp->sk_refcnt), sp);
920 }
921 
922 int udp6_seq_show(struct seq_file *seq, void *v)
923 {
924 	if (v == SEQ_START_TOKEN)
925 		seq_printf(seq,
926 			   "  sl  "
927 			   "local_address                         "
928 			   "remote_address                        "
929 			   "st tx_queue rx_queue tr tm->when retrnsmt"
930 			   "   uid  timeout inode\n");
931 	else
932 		udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
933 	return 0;
934 }
935 
936 static struct file_operations udp6_seq_fops;
937 static struct udp_seq_afinfo udp6_seq_afinfo = {
938 	.owner		= THIS_MODULE,
939 	.name		= "udp6",
940 	.family		= AF_INET6,
941 	.hashtable	= udp_hash,
942 	.seq_show	= udp6_seq_show,
943 	.seq_fops	= &udp6_seq_fops,
944 };
945 
946 int __init udp6_proc_init(void)
947 {
948 	return udp_proc_register(&udp6_seq_afinfo);
949 }
950 
951 void udp6_proc_exit(void) {
952 	udp_proc_unregister(&udp6_seq_afinfo);
953 }
954 #endif /* CONFIG_PROC_FS */
955 
956 /* ------------------------------------------------------------------------ */
957 
958 struct proto udpv6_prot = {
959 	.name		   = "UDPv6",
960 	.owner		   = THIS_MODULE,
961 	.close		   = udp_lib_close,
962 	.connect	   = ip6_datagram_connect,
963 	.disconnect	   = udp_disconnect,
964 	.ioctl		   = udp_ioctl,
965 	.destroy	   = udpv6_destroy_sock,
966 	.setsockopt	   = udpv6_setsockopt,
967 	.getsockopt	   = udpv6_getsockopt,
968 	.sendmsg	   = udpv6_sendmsg,
969 	.recvmsg	   = udpv6_recvmsg,
970 	.backlog_rcv	   = udpv6_queue_rcv_skb,
971 	.hash		   = udp_lib_hash,
972 	.unhash		   = udp_lib_unhash,
973 	.get_port	   = udp_v6_get_port,
974 	.obj_size	   = sizeof(struct udp6_sock),
975 #ifdef CONFIG_COMPAT
976 	.compat_setsockopt = compat_udpv6_setsockopt,
977 	.compat_getsockopt = compat_udpv6_getsockopt,
978 #endif
979 };
980 
981 static struct inet_protosw udpv6_protosw = {
982 	.type =      SOCK_DGRAM,
983 	.protocol =  IPPROTO_UDP,
984 	.prot =      &udpv6_prot,
985 	.ops =       &inet6_dgram_ops,
986 	.capability =-1,
987 	.no_check =  UDP_CSUM_DEFAULT,
988 	.flags =     INET_PROTOSW_PERMANENT,
989 };
990 
991 
992 void __init udpv6_init(void)
993 {
994 	if (inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP) < 0)
995 		printk(KERN_ERR "udpv6_init: Could not register protocol\n");
996 	inet6_register_protosw(&udpv6_protosw);
997 }
998