xref: /openbmc/linux/net/ipv6/udp.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
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 int ipv6_rcv_saddr_any(const struct sock *sk)
56 {
57 	struct ipv6_pinfo *np = inet6_sk(sk);
58 
59 	return ipv6_addr_any(&np->rcv_saddr);
60 }
61 
62 static unsigned int ipv6_hash_port_and_rcv_saddr(__u16 port,
63 						 const struct sock *sk)
64 {
65 	return port;
66 }
67 
68 const struct udp_get_port_ops udp_ipv6_ops = {
69 	.saddr_cmp = ipv6_rcv_saddr_equal,
70 	.saddr_any = ipv6_rcv_saddr_any,
71 	.hash_port_and_rcv_saddr = ipv6_hash_port_and_rcv_saddr,
72 };
73 
74 static inline int udp_v6_get_port(struct sock *sk, unsigned short snum)
75 {
76 	return udp_get_port(sk, snum, &udp_ipv6_ops);
77 }
78 
79 static struct sock *__udp6_lib_lookup(struct in6_addr *saddr, __be16 sport,
80 				      struct in6_addr *daddr, __be16 dport,
81 				      int dif, struct hlist_head udptable[])
82 {
83 	struct sock *sk, *result = NULL;
84 	struct hlist_node *node;
85 	unsigned short hnum = ntohs(dport);
86 	int badness = -1;
87 
88 	read_lock(&udp_hash_lock);
89 	sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
90 		struct inet_sock *inet = inet_sk(sk);
91 
92 		if (sk->sk_hash == hnum && sk->sk_family == PF_INET6) {
93 			struct ipv6_pinfo *np = inet6_sk(sk);
94 			int score = 0;
95 			if (inet->dport) {
96 				if (inet->dport != sport)
97 					continue;
98 				score++;
99 			}
100 			if (!ipv6_addr_any(&np->rcv_saddr)) {
101 				if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
102 					continue;
103 				score++;
104 			}
105 			if (!ipv6_addr_any(&np->daddr)) {
106 				if (!ipv6_addr_equal(&np->daddr, saddr))
107 					continue;
108 				score++;
109 			}
110 			if (sk->sk_bound_dev_if) {
111 				if (sk->sk_bound_dev_if != dif)
112 					continue;
113 				score++;
114 			}
115 			if (score == 4) {
116 				result = sk;
117 				break;
118 			} else if (score > badness) {
119 				result = sk;
120 				badness = score;
121 			}
122 		}
123 	}
124 	if (result)
125 		sock_hold(result);
126 	read_unlock(&udp_hash_lock);
127 	return result;
128 }
129 
130 /*
131  * 	This should be easy, if there is something there we
132  * 	return it, otherwise we block.
133  */
134 
135 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
136 		  struct msghdr *msg, size_t len,
137 		  int noblock, int flags, int *addr_len)
138 {
139 	struct ipv6_pinfo *np = inet6_sk(sk);
140 	struct inet_sock *inet = inet_sk(sk);
141 	struct sk_buff *skb;
142 	unsigned int ulen, copied;
143 	int err;
144 	int is_udplite = IS_UDPLITE(sk);
145 
146 	if (addr_len)
147 		*addr_len=sizeof(struct sockaddr_in6);
148 
149 	if (flags & MSG_ERRQUEUE)
150 		return ipv6_recv_error(sk, msg, len);
151 
152 try_again:
153 	skb = skb_recv_datagram(sk, flags, noblock, &err);
154 	if (!skb)
155 		goto out;
156 
157 	ulen = skb->len - sizeof(struct udphdr);
158 	copied = len;
159 	if (copied > ulen)
160 		copied = ulen;
161 	else if (copied < ulen)
162 		msg->msg_flags |= MSG_TRUNC;
163 
164 	/*
165 	 * If checksum is needed at all, try to do it while copying the
166 	 * data.  If the data is truncated, or if we only want a partial
167 	 * coverage checksum (UDP-Lite), do it before the copy.
168 	 */
169 
170 	if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
171 		if (udp_lib_checksum_complete(skb))
172 			goto csum_copy_err;
173 	}
174 
175 	if (skb_csum_unnecessary(skb))
176 		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
177 					      msg->msg_iov, copied       );
178 	else {
179 		err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
180 		if (err == -EINVAL)
181 			goto csum_copy_err;
182 	}
183 	if (err)
184 		goto out_free;
185 
186 	sock_recv_timestamp(msg, sk, skb);
187 
188 	/* Copy the address. */
189 	if (msg->msg_name) {
190 		struct sockaddr_in6 *sin6;
191 
192 		sin6 = (struct sockaddr_in6 *) msg->msg_name;
193 		sin6->sin6_family = AF_INET6;
194 		sin6->sin6_port = udp_hdr(skb)->source;
195 		sin6->sin6_flowinfo = 0;
196 		sin6->sin6_scope_id = 0;
197 
198 		if (skb->protocol == htons(ETH_P_IP))
199 			ipv6_addr_set(&sin6->sin6_addr, 0, 0,
200 				      htonl(0xffff), ip_hdr(skb)->saddr);
201 		else {
202 			ipv6_addr_copy(&sin6->sin6_addr,
203 				       &ipv6_hdr(skb)->saddr);
204 			if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
205 				sin6->sin6_scope_id = IP6CB(skb)->iif;
206 		}
207 
208 	}
209 	if (skb->protocol == htons(ETH_P_IP)) {
210 		if (inet->cmsg_flags)
211 			ip_cmsg_recv(msg, skb);
212 	} else {
213 		if (np->rxopt.all)
214 			datagram_recv_ctl(sk, msg, skb);
215 	}
216 
217 	err = copied;
218 	if (flags & MSG_TRUNC)
219 		err = ulen;
220 
221 out_free:
222 	skb_free_datagram(sk, skb);
223 out:
224 	return err;
225 
226 csum_copy_err:
227 	skb_kill_datagram(sk, skb, flags);
228 
229 	if (flags & MSG_DONTWAIT) {
230 		UDP6_INC_STATS_USER(UDP_MIB_INERRORS, is_udplite);
231 		return -EAGAIN;
232 	}
233 	goto try_again;
234 }
235 
236 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
237 		    int type, int code, int offset, __be32 info,
238 		    struct hlist_head udptable[]                    )
239 {
240 	struct ipv6_pinfo *np;
241 	struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
242 	struct in6_addr *saddr = &hdr->saddr;
243 	struct in6_addr *daddr = &hdr->daddr;
244 	struct udphdr *uh = (struct udphdr*)(skb->data+offset);
245 	struct sock *sk;
246 	int err;
247 
248 	sk = __udp6_lib_lookup(daddr, uh->dest,
249 			       saddr, uh->source, inet6_iif(skb), udptable);
250 	if (sk == NULL)
251 		return;
252 
253 	np = inet6_sk(sk);
254 
255 	if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
256 		goto out;
257 
258 	if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
259 		goto out;
260 
261 	if (np->recverr)
262 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
263 
264 	sk->sk_err = err;
265 	sk->sk_error_report(sk);
266 out:
267 	sock_put(sk);
268 }
269 
270 static __inline__ void udpv6_err(struct sk_buff *skb,
271 				 struct inet6_skb_parm *opt, int type,
272 				 int code, int offset, __be32 info     )
273 {
274 	return __udp6_lib_err(skb, opt, type, code, offset, info, udp_hash);
275 }
276 
277 int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
278 {
279 	struct udp_sock *up = udp_sk(sk);
280 	int rc;
281 
282 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
283 		goto drop;
284 
285 	/*
286 	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
287 	 */
288 	if ((up->pcflag & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
289 
290 		if (up->pcrlen == 0) {          /* full coverage was set  */
291 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
292 				" %d while full coverage %d requested\n",
293 				UDP_SKB_CB(skb)->cscov, skb->len);
294 			goto drop;
295 		}
296 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
297 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
298 						    "too small, need min %d\n",
299 				       UDP_SKB_CB(skb)->cscov, up->pcrlen);
300 			goto drop;
301 		}
302 	}
303 
304 	if (sk->sk_filter) {
305 		if (udp_lib_checksum_complete(skb))
306 			goto drop;
307 	}
308 
309 	if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
310 		/* Note that an ENOMEM error is charged twice */
311 		if (rc == -ENOMEM)
312 			UDP6_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, up->pcflag);
313 		goto drop;
314 	}
315 	UDP6_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
316 	return 0;
317 drop:
318 	UDP6_INC_STATS_BH(UDP_MIB_INERRORS, up->pcflag);
319 	kfree_skb(skb);
320 	return -1;
321 }
322 
323 static struct sock *udp_v6_mcast_next(struct sock *sk,
324 				      __be16 loc_port, struct in6_addr *loc_addr,
325 				      __be16 rmt_port, struct in6_addr *rmt_addr,
326 				      int dif)
327 {
328 	struct hlist_node *node;
329 	struct sock *s = sk;
330 	unsigned short num = ntohs(loc_port);
331 
332 	sk_for_each_from(s, node) {
333 		struct inet_sock *inet = inet_sk(s);
334 
335 		if (s->sk_hash == num && s->sk_family == PF_INET6) {
336 			struct ipv6_pinfo *np = inet6_sk(s);
337 			if (inet->dport) {
338 				if (inet->dport != rmt_port)
339 					continue;
340 			}
341 			if (!ipv6_addr_any(&np->daddr) &&
342 			    !ipv6_addr_equal(&np->daddr, rmt_addr))
343 				continue;
344 
345 			if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
346 				continue;
347 
348 			if (!ipv6_addr_any(&np->rcv_saddr)) {
349 				if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
350 					continue;
351 			}
352 			if (!inet6_mc_check(s, loc_addr, rmt_addr))
353 				continue;
354 			return s;
355 		}
356 	}
357 	return NULL;
358 }
359 
360 /*
361  * Note: called only from the BH handler context,
362  * so we don't need to lock the hashes.
363  */
364 static int __udp6_lib_mcast_deliver(struct sk_buff *skb, struct in6_addr *saddr,
365 			   struct in6_addr *daddr, struct hlist_head udptable[])
366 {
367 	struct sock *sk, *sk2;
368 	const struct udphdr *uh = udp_hdr(skb);
369 	int dif;
370 
371 	read_lock(&udp_hash_lock);
372 	sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
373 	dif = inet6_iif(skb);
374 	sk = udp_v6_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
375 	if (!sk) {
376 		kfree_skb(skb);
377 		goto out;
378 	}
379 
380 	sk2 = sk;
381 	while ((sk2 = udp_v6_mcast_next(sk_next(sk2), uh->dest, daddr,
382 					uh->source, saddr, dif))) {
383 		struct sk_buff *buff = skb_clone(skb, GFP_ATOMIC);
384 		if (buff)
385 			udpv6_queue_rcv_skb(sk2, buff);
386 	}
387 	udpv6_queue_rcv_skb(sk, skb);
388 out:
389 	read_unlock(&udp_hash_lock);
390 	return 0;
391 }
392 
393 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
394 				 int proto)
395 {
396 	int err;
397 
398 	UDP_SKB_CB(skb)->partial_cov = 0;
399 	UDP_SKB_CB(skb)->cscov = skb->len;
400 
401 	if (proto == IPPROTO_UDPLITE) {
402 		err = udplite_checksum_init(skb, uh);
403 		if (err)
404 			return err;
405 	}
406 
407 	if (uh->check == 0) {
408 		/* RFC 2460 section 8.1 says that we SHOULD log
409 		   this error. Well, it is reasonable.
410 		 */
411 		LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
412 		return 1;
413 	}
414 	if (skb->ip_summed == CHECKSUM_COMPLETE &&
415 	    !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
416 			     skb->len, proto, skb->csum))
417 		skb->ip_summed = CHECKSUM_UNNECESSARY;
418 
419 	if (!skb_csum_unnecessary(skb))
420 		skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
421 							 &ipv6_hdr(skb)->daddr,
422 							 skb->len, proto, 0));
423 
424 	return 0;
425 }
426 
427 int __udp6_lib_rcv(struct sk_buff **pskb, struct hlist_head udptable[],
428 		   int proto)
429 {
430 	struct sk_buff *skb = *pskb;
431 	struct sock *sk;
432 	struct udphdr *uh;
433 	struct net_device *dev = skb->dev;
434 	struct in6_addr *saddr, *daddr;
435 	u32 ulen = 0;
436 
437 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
438 		goto short_packet;
439 
440 	saddr = &ipv6_hdr(skb)->saddr;
441 	daddr = &ipv6_hdr(skb)->daddr;
442 	uh = udp_hdr(skb);
443 
444 	ulen = ntohs(uh->len);
445 	if (ulen > skb->len)
446 		goto short_packet;
447 
448 	if (proto == IPPROTO_UDP) {
449 		/* UDP validates ulen. */
450 
451 		/* Check for jumbo payload */
452 		if (ulen == 0)
453 			ulen = skb->len;
454 
455 		if (ulen < sizeof(*uh))
456 			goto short_packet;
457 
458 		if (ulen < skb->len) {
459 			if (pskb_trim_rcsum(skb, ulen))
460 				goto short_packet;
461 			saddr = &ipv6_hdr(skb)->saddr;
462 			daddr = &ipv6_hdr(skb)->daddr;
463 			uh = udp_hdr(skb);
464 		}
465 	}
466 
467 	if (udp6_csum_init(skb, uh, proto))
468 		goto discard;
469 
470 	/*
471 	 *	Multicast receive code
472 	 */
473 	if (ipv6_addr_is_multicast(daddr))
474 		return __udp6_lib_mcast_deliver(skb, saddr, daddr, udptable);
475 
476 	/* Unicast */
477 
478 	/*
479 	 * check socket cache ... must talk to Alan about his plans
480 	 * for sock caches... i'll skip this for now.
481 	 */
482 	sk = __udp6_lib_lookup(saddr, uh->source,
483 			       daddr, uh->dest, inet6_iif(skb), udptable);
484 
485 	if (sk == NULL) {
486 		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
487 			goto discard;
488 
489 		if (udp_lib_checksum_complete(skb))
490 			goto discard;
491 		UDP6_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
492 
493 		icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0, dev);
494 
495 		kfree_skb(skb);
496 		return 0;
497 	}
498 
499 	/* deliver */
500 
501 	udpv6_queue_rcv_skb(sk, skb);
502 	sock_put(sk);
503 	return 0;
504 
505 short_packet:
506 	LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: %d/%u\n",
507 		       proto == IPPROTO_UDPLITE ? "-Lite" : "",
508 		       ulen, skb->len);
509 
510 discard:
511 	UDP6_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
512 	kfree_skb(skb);
513 	return 0;
514 }
515 
516 static __inline__ int udpv6_rcv(struct sk_buff **pskb)
517 {
518 	return __udp6_lib_rcv(pskb, udp_hash, IPPROTO_UDP);
519 }
520 
521 /*
522  * Throw away all pending data and cancel the corking. Socket is locked.
523  */
524 static void udp_v6_flush_pending_frames(struct sock *sk)
525 {
526 	struct udp_sock *up = udp_sk(sk);
527 
528 	if (up->pending) {
529 		up->len = 0;
530 		up->pending = 0;
531 		ip6_flush_pending_frames(sk);
532 	}
533 }
534 
535 /*
536  *	Sending
537  */
538 
539 static int udp_v6_push_pending_frames(struct sock *sk)
540 {
541 	struct sk_buff *skb;
542 	struct udphdr *uh;
543 	struct udp_sock  *up = udp_sk(sk);
544 	struct inet_sock *inet = inet_sk(sk);
545 	struct flowi *fl = &inet->cork.fl;
546 	int err = 0;
547 	__wsum csum = 0;
548 
549 	/* Grab the skbuff where UDP header space exists. */
550 	if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
551 		goto out;
552 
553 	/*
554 	 * Create a UDP header
555 	 */
556 	uh = udp_hdr(skb);
557 	uh->source = fl->fl_ip_sport;
558 	uh->dest = fl->fl_ip_dport;
559 	uh->len = htons(up->len);
560 	uh->check = 0;
561 
562 	if (up->pcflag)
563 		csum = udplite_csum_outgoing(sk, skb);
564 	 else
565 		csum = udp_csum_outgoing(sk, skb);
566 
567 	/* add protocol-dependent pseudo-header */
568 	uh->check = csum_ipv6_magic(&fl->fl6_src, &fl->fl6_dst,
569 				    up->len, fl->proto, csum   );
570 	if (uh->check == 0)
571 		uh->check = CSUM_MANGLED_0;
572 
573 	err = ip6_push_pending_frames(sk);
574 out:
575 	up->len = 0;
576 	up->pending = 0;
577 	return err;
578 }
579 
580 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
581 		  struct msghdr *msg, size_t len)
582 {
583 	struct ipv6_txoptions opt_space;
584 	struct udp_sock *up = udp_sk(sk);
585 	struct inet_sock *inet = inet_sk(sk);
586 	struct ipv6_pinfo *np = inet6_sk(sk);
587 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
588 	struct in6_addr *daddr, *final_p = NULL, final;
589 	struct ipv6_txoptions *opt = NULL;
590 	struct ip6_flowlabel *flowlabel = NULL;
591 	struct flowi fl;
592 	struct dst_entry *dst;
593 	int addr_len = msg->msg_namelen;
594 	int ulen = len;
595 	int hlimit = -1;
596 	int tclass = -1;
597 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
598 	int err;
599 	int connected = 0;
600 	int is_udplite = up->pcflag;
601 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
602 
603 	/* destination address check */
604 	if (sin6) {
605 		if (addr_len < offsetof(struct sockaddr, sa_data))
606 			return -EINVAL;
607 
608 		switch (sin6->sin6_family) {
609 		case AF_INET6:
610 			if (addr_len < SIN6_LEN_RFC2133)
611 				return -EINVAL;
612 			daddr = &sin6->sin6_addr;
613 			break;
614 		case AF_INET:
615 			goto do_udp_sendmsg;
616 		case AF_UNSPEC:
617 			msg->msg_name = sin6 = NULL;
618 			msg->msg_namelen = addr_len = 0;
619 			daddr = NULL;
620 			break;
621 		default:
622 			return -EINVAL;
623 		}
624 	} else if (!up->pending) {
625 		if (sk->sk_state != TCP_ESTABLISHED)
626 			return -EDESTADDRREQ;
627 		daddr = &np->daddr;
628 	} else
629 		daddr = NULL;
630 
631 	if (daddr) {
632 		if (ipv6_addr_type(daddr) == IPV6_ADDR_MAPPED) {
633 			struct sockaddr_in sin;
634 			sin.sin_family = AF_INET;
635 			sin.sin_port = sin6 ? sin6->sin6_port : inet->dport;
636 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
637 			msg->msg_name = &sin;
638 			msg->msg_namelen = sizeof(sin);
639 do_udp_sendmsg:
640 			if (__ipv6_only_sock(sk))
641 				return -ENETUNREACH;
642 			return udp_sendmsg(iocb, sk, msg, len);
643 		}
644 	}
645 
646 	if (up->pending == AF_INET)
647 		return udp_sendmsg(iocb, sk, msg, len);
648 
649 	/* Rough check on arithmetic overflow,
650 	   better check is made in ip6_append_data().
651 	   */
652 	if (len > INT_MAX - sizeof(struct udphdr))
653 		return -EMSGSIZE;
654 
655 	if (up->pending) {
656 		/*
657 		 * There are pending frames.
658 		 * The socket lock must be held while it's corked.
659 		 */
660 		lock_sock(sk);
661 		if (likely(up->pending)) {
662 			if (unlikely(up->pending != AF_INET6)) {
663 				release_sock(sk);
664 				return -EAFNOSUPPORT;
665 			}
666 			dst = NULL;
667 			goto do_append_data;
668 		}
669 		release_sock(sk);
670 	}
671 	ulen += sizeof(struct udphdr);
672 
673 	memset(&fl, 0, sizeof(fl));
674 
675 	if (sin6) {
676 		if (sin6->sin6_port == 0)
677 			return -EINVAL;
678 
679 		fl.fl_ip_dport = sin6->sin6_port;
680 		daddr = &sin6->sin6_addr;
681 
682 		if (np->sndflow) {
683 			fl.fl6_flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
684 			if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
685 				flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
686 				if (flowlabel == NULL)
687 					return -EINVAL;
688 				daddr = &flowlabel->dst;
689 			}
690 		}
691 
692 		/*
693 		 * Otherwise it will be difficult to maintain
694 		 * sk->sk_dst_cache.
695 		 */
696 		if (sk->sk_state == TCP_ESTABLISHED &&
697 		    ipv6_addr_equal(daddr, &np->daddr))
698 			daddr = &np->daddr;
699 
700 		if (addr_len >= sizeof(struct sockaddr_in6) &&
701 		    sin6->sin6_scope_id &&
702 		    ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
703 			fl.oif = sin6->sin6_scope_id;
704 	} else {
705 		if (sk->sk_state != TCP_ESTABLISHED)
706 			return -EDESTADDRREQ;
707 
708 		fl.fl_ip_dport = inet->dport;
709 		daddr = &np->daddr;
710 		fl.fl6_flowlabel = np->flow_label;
711 		connected = 1;
712 	}
713 
714 	if (!fl.oif)
715 		fl.oif = sk->sk_bound_dev_if;
716 
717 	if (msg->msg_controllen) {
718 		opt = &opt_space;
719 		memset(opt, 0, sizeof(struct ipv6_txoptions));
720 		opt->tot_len = sizeof(*opt);
721 
722 		err = datagram_send_ctl(msg, &fl, opt, &hlimit, &tclass);
723 		if (err < 0) {
724 			fl6_sock_release(flowlabel);
725 			return err;
726 		}
727 		if ((fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
728 			flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
729 			if (flowlabel == NULL)
730 				return -EINVAL;
731 		}
732 		if (!(opt->opt_nflen|opt->opt_flen))
733 			opt = NULL;
734 		connected = 0;
735 	}
736 	if (opt == NULL)
737 		opt = np->opt;
738 	if (flowlabel)
739 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
740 	opt = ipv6_fixup_options(&opt_space, opt);
741 
742 	fl.proto = sk->sk_protocol;
743 	ipv6_addr_copy(&fl.fl6_dst, daddr);
744 	if (ipv6_addr_any(&fl.fl6_src) && !ipv6_addr_any(&np->saddr))
745 		ipv6_addr_copy(&fl.fl6_src, &np->saddr);
746 	fl.fl_ip_sport = inet->sport;
747 
748 	/* merge ip6_build_xmit from ip6_output */
749 	if (opt && opt->srcrt) {
750 		struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
751 		ipv6_addr_copy(&final, &fl.fl6_dst);
752 		ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
753 		final_p = &final;
754 		connected = 0;
755 	}
756 
757 	if (!fl.oif && ipv6_addr_is_multicast(&fl.fl6_dst)) {
758 		fl.oif = np->mcast_oif;
759 		connected = 0;
760 	}
761 
762 	security_sk_classify_flow(sk, &fl);
763 
764 	err = ip6_sk_dst_lookup(sk, &dst, &fl);
765 	if (err)
766 		goto out;
767 	if (final_p)
768 		ipv6_addr_copy(&fl.fl6_dst, final_p);
769 
770 	if ((err = xfrm_lookup(&dst, &fl, sk, 1)) < 0)
771 		goto out;
772 
773 	if (hlimit < 0) {
774 		if (ipv6_addr_is_multicast(&fl.fl6_dst))
775 			hlimit = np->mcast_hops;
776 		else
777 			hlimit = np->hop_limit;
778 		if (hlimit < 0)
779 			hlimit = dst_metric(dst, RTAX_HOPLIMIT);
780 		if (hlimit < 0)
781 			hlimit = ipv6_get_hoplimit(dst->dev);
782 	}
783 
784 	if (tclass < 0) {
785 		tclass = np->tclass;
786 		if (tclass < 0)
787 			tclass = 0;
788 	}
789 
790 	if (msg->msg_flags&MSG_CONFIRM)
791 		goto do_confirm;
792 back_from_confirm:
793 
794 	lock_sock(sk);
795 	if (unlikely(up->pending)) {
796 		/* The socket is already corked while preparing it. */
797 		/* ... which is an evident application bug. --ANK */
798 		release_sock(sk);
799 
800 		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
801 		err = -EINVAL;
802 		goto out;
803 	}
804 
805 	up->pending = AF_INET6;
806 
807 do_append_data:
808 	up->len += ulen;
809 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
810 	err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
811 		sizeof(struct udphdr), hlimit, tclass, opt, &fl,
812 		(struct rt6_info*)dst,
813 		corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
814 	if (err)
815 		udp_v6_flush_pending_frames(sk);
816 	else if (!corkreq)
817 		err = udp_v6_push_pending_frames(sk);
818 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
819 		up->pending = 0;
820 
821 	if (dst) {
822 		if (connected) {
823 			ip6_dst_store(sk, dst,
824 				      ipv6_addr_equal(&fl.fl6_dst, &np->daddr) ?
825 				      &np->daddr : NULL,
826 #ifdef CONFIG_IPV6_SUBTREES
827 				      ipv6_addr_equal(&fl.fl6_src, &np->saddr) ?
828 				      &np->saddr :
829 #endif
830 				      NULL);
831 		} else {
832 			dst_release(dst);
833 		}
834 	}
835 
836 	if (err > 0)
837 		err = np->recverr ? net_xmit_errno(err) : 0;
838 	release_sock(sk);
839 out:
840 	fl6_sock_release(flowlabel);
841 	if (!err) {
842 		UDP6_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
843 		return len;
844 	}
845 	/*
846 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
847 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
848 	 * we don't have a good statistic (IpOutDiscards but it can be too many
849 	 * things).  We could add another new stat but at least for now that
850 	 * seems like overkill.
851 	 */
852 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
853 		UDP6_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
854 	}
855 	return err;
856 
857 do_confirm:
858 	dst_confirm(dst);
859 	if (!(msg->msg_flags&MSG_PROBE) || len)
860 		goto back_from_confirm;
861 	err = 0;
862 	goto out;
863 }
864 
865 int udpv6_destroy_sock(struct sock *sk)
866 {
867 	lock_sock(sk);
868 	udp_v6_flush_pending_frames(sk);
869 	release_sock(sk);
870 
871 	inet6_destroy_sock(sk);
872 
873 	return 0;
874 }
875 
876 /*
877  *	Socket option code for UDP
878  */
879 int udpv6_setsockopt(struct sock *sk, int level, int optname,
880 		     char __user *optval, int optlen)
881 {
882 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
883 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
884 					  udp_v6_push_pending_frames);
885 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
886 }
887 
888 #ifdef CONFIG_COMPAT
889 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
890 			    char __user *optval, int optlen)
891 {
892 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
893 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
894 					  udp_v6_push_pending_frames);
895 	return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
896 }
897 #endif
898 
899 int udpv6_getsockopt(struct sock *sk, int level, int optname,
900 		     char __user *optval, int __user *optlen)
901 {
902 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
903 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
904 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
905 }
906 
907 #ifdef CONFIG_COMPAT
908 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
909 			    char __user *optval, int __user *optlen)
910 {
911 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
912 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
913 	return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
914 }
915 #endif
916 
917 static struct inet6_protocol udpv6_protocol = {
918 	.handler	=	udpv6_rcv,
919 	.err_handler	=	udpv6_err,
920 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
921 };
922 
923 /* ------------------------------------------------------------------------ */
924 #ifdef CONFIG_PROC_FS
925 
926 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
927 {
928 	struct inet_sock *inet = inet_sk(sp);
929 	struct ipv6_pinfo *np = inet6_sk(sp);
930 	struct in6_addr *dest, *src;
931 	__u16 destp, srcp;
932 
933 	dest  = &np->daddr;
934 	src   = &np->rcv_saddr;
935 	destp = ntohs(inet->dport);
936 	srcp  = ntohs(inet->sport);
937 	seq_printf(seq,
938 		   "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
939 		   "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p\n",
940 		   bucket,
941 		   src->s6_addr32[0], src->s6_addr32[1],
942 		   src->s6_addr32[2], src->s6_addr32[3], srcp,
943 		   dest->s6_addr32[0], dest->s6_addr32[1],
944 		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
945 		   sp->sk_state,
946 		   atomic_read(&sp->sk_wmem_alloc),
947 		   atomic_read(&sp->sk_rmem_alloc),
948 		   0, 0L, 0,
949 		   sock_i_uid(sp), 0,
950 		   sock_i_ino(sp),
951 		   atomic_read(&sp->sk_refcnt), sp);
952 }
953 
954 int udp6_seq_show(struct seq_file *seq, void *v)
955 {
956 	if (v == SEQ_START_TOKEN)
957 		seq_printf(seq,
958 			   "  sl  "
959 			   "local_address                         "
960 			   "remote_address                        "
961 			   "st tx_queue rx_queue tr tm->when retrnsmt"
962 			   "   uid  timeout inode\n");
963 	else
964 		udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
965 	return 0;
966 }
967 
968 static struct file_operations udp6_seq_fops;
969 static struct udp_seq_afinfo udp6_seq_afinfo = {
970 	.owner		= THIS_MODULE,
971 	.name		= "udp6",
972 	.family		= AF_INET6,
973 	.hashtable	= udp_hash,
974 	.seq_show	= udp6_seq_show,
975 	.seq_fops	= &udp6_seq_fops,
976 };
977 
978 int __init udp6_proc_init(void)
979 {
980 	return udp_proc_register(&udp6_seq_afinfo);
981 }
982 
983 void udp6_proc_exit(void) {
984 	udp_proc_unregister(&udp6_seq_afinfo);
985 }
986 #endif /* CONFIG_PROC_FS */
987 
988 /* ------------------------------------------------------------------------ */
989 
990 struct proto udpv6_prot = {
991 	.name		   = "UDPv6",
992 	.owner		   = THIS_MODULE,
993 	.close		   = udp_lib_close,
994 	.connect	   = ip6_datagram_connect,
995 	.disconnect	   = udp_disconnect,
996 	.ioctl		   = udp_ioctl,
997 	.destroy	   = udpv6_destroy_sock,
998 	.setsockopt	   = udpv6_setsockopt,
999 	.getsockopt	   = udpv6_getsockopt,
1000 	.sendmsg	   = udpv6_sendmsg,
1001 	.recvmsg	   = udpv6_recvmsg,
1002 	.backlog_rcv	   = udpv6_queue_rcv_skb,
1003 	.hash		   = udp_lib_hash,
1004 	.unhash		   = udp_lib_unhash,
1005 	.get_port	   = udp_v6_get_port,
1006 	.obj_size	   = sizeof(struct udp6_sock),
1007 #ifdef CONFIG_COMPAT
1008 	.compat_setsockopt = compat_udpv6_setsockopt,
1009 	.compat_getsockopt = compat_udpv6_getsockopt,
1010 #endif
1011 };
1012 
1013 static struct inet_protosw udpv6_protosw = {
1014 	.type =      SOCK_DGRAM,
1015 	.protocol =  IPPROTO_UDP,
1016 	.prot =      &udpv6_prot,
1017 	.ops =       &inet6_dgram_ops,
1018 	.capability =-1,
1019 	.no_check =  UDP_CSUM_DEFAULT,
1020 	.flags =     INET_PROTOSW_PERMANENT,
1021 };
1022 
1023 
1024 void __init udpv6_init(void)
1025 {
1026 	if (inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP) < 0)
1027 		printk(KERN_ERR "udpv6_init: Could not register protocol\n");
1028 	inet6_register_protosw(&udpv6_protosw);
1029 }
1030