xref: /linux/net/ipv6/udp.c (revision b8bb76713ec50df2f11efee386e16f93d51e1076)
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 ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
53 {
54 	const struct in6_addr *sk_rcv_saddr6 = &inet6_sk(sk)->rcv_saddr;
55 	const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
56 	int sk_ipv6only = ipv6_only_sock(sk);
57 	int sk2_ipv6only = inet_v6_ipv6only(sk2);
58 	int addr_type = ipv6_addr_type(sk_rcv_saddr6);
59 	int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
60 
61 	/* if both are mapped, treat as IPv4 */
62 	if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
63 		return ipv4_rcv_saddr_equal(sk, sk2);
64 
65 	if (addr_type2 == IPV6_ADDR_ANY &&
66 	    !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
67 		return 1;
68 
69 	if (addr_type == IPV6_ADDR_ANY &&
70 	    !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
71 		return 1;
72 
73 	if (sk2_rcv_saddr6 &&
74 	    ipv6_addr_equal(sk_rcv_saddr6, sk2_rcv_saddr6))
75 		return 1;
76 
77 	return 0;
78 }
79 
80 int udp_v6_get_port(struct sock *sk, unsigned short snum)
81 {
82 	return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal);
83 }
84 
85 static inline int compute_score(struct sock *sk, struct net *net,
86 				unsigned short hnum,
87 				struct in6_addr *saddr, __be16 sport,
88 				struct in6_addr *daddr, __be16 dport,
89 				int dif)
90 {
91 	int score = -1;
92 
93 	if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum &&
94 			sk->sk_family == PF_INET6) {
95 		struct ipv6_pinfo *np = inet6_sk(sk);
96 		struct inet_sock *inet = inet_sk(sk);
97 
98 		score = 0;
99 		if (inet->dport) {
100 			if (inet->dport != sport)
101 				return -1;
102 			score++;
103 		}
104 		if (!ipv6_addr_any(&np->rcv_saddr)) {
105 			if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
106 				return -1;
107 			score++;
108 		}
109 		if (!ipv6_addr_any(&np->daddr)) {
110 			if (!ipv6_addr_equal(&np->daddr, saddr))
111 				return -1;
112 			score++;
113 		}
114 		if (sk->sk_bound_dev_if) {
115 			if (sk->sk_bound_dev_if != dif)
116 				return -1;
117 			score++;
118 		}
119 	}
120 	return score;
121 }
122 
123 static struct sock *__udp6_lib_lookup(struct net *net,
124 				      struct in6_addr *saddr, __be16 sport,
125 				      struct in6_addr *daddr, __be16 dport,
126 				      int dif, struct udp_table *udptable)
127 {
128 	struct sock *sk, *result;
129 	struct hlist_nulls_node *node;
130 	unsigned short hnum = ntohs(dport);
131 	unsigned int hash = udp_hashfn(net, hnum);
132 	struct udp_hslot *hslot = &udptable->hash[hash];
133 	int score, badness;
134 
135 	rcu_read_lock();
136 begin:
137 	result = NULL;
138 	badness = -1;
139 	sk_nulls_for_each_rcu(sk, node, &hslot->head) {
140 		score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
141 		if (score > badness) {
142 			result = sk;
143 			badness = score;
144 		}
145 	}
146 	/*
147 	 * if the nulls value we got at the end of this lookup is
148 	 * not the expected one, we must restart lookup.
149 	 * We probably met an item that was moved to another chain.
150 	 */
151 	if (get_nulls_value(node) != hash)
152 		goto begin;
153 
154 	if (result) {
155 		if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
156 			result = NULL;
157 		else if (unlikely(compute_score(result, net, hnum, saddr, sport,
158 					daddr, dport, dif) < badness)) {
159 			sock_put(result);
160 			goto begin;
161 		}
162 	}
163 	rcu_read_unlock();
164 	return result;
165 }
166 
167 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
168 					  __be16 sport, __be16 dport,
169 					  struct udp_table *udptable)
170 {
171 	struct sock *sk;
172 	struct ipv6hdr *iph = ipv6_hdr(skb);
173 
174 	if (unlikely(sk = skb_steal_sock(skb)))
175 		return sk;
176 	else
177 		return __udp6_lib_lookup(dev_net(skb->dst->dev), &iph->saddr, sport,
178 					 &iph->daddr, dport, inet6_iif(skb),
179 					 udptable);
180 }
181 
182 /*
183  * 	This should be easy, if there is something there we
184  * 	return it, otherwise we block.
185  */
186 
187 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
188 		  struct msghdr *msg, size_t len,
189 		  int noblock, int flags, int *addr_len)
190 {
191 	struct ipv6_pinfo *np = inet6_sk(sk);
192 	struct inet_sock *inet = inet_sk(sk);
193 	struct sk_buff *skb;
194 	unsigned int ulen, copied;
195 	int peeked;
196 	int err;
197 	int is_udplite = IS_UDPLITE(sk);
198 	int is_udp4;
199 
200 	if (addr_len)
201 		*addr_len=sizeof(struct sockaddr_in6);
202 
203 	if (flags & MSG_ERRQUEUE)
204 		return ipv6_recv_error(sk, msg, len);
205 
206 try_again:
207 	skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
208 				  &peeked, &err);
209 	if (!skb)
210 		goto out;
211 
212 	ulen = skb->len - sizeof(struct udphdr);
213 	copied = len;
214 	if (copied > ulen)
215 		copied = ulen;
216 	else if (copied < ulen)
217 		msg->msg_flags |= MSG_TRUNC;
218 
219 	is_udp4 = (skb->protocol == htons(ETH_P_IP));
220 
221 	/*
222 	 * If checksum is needed at all, try to do it while copying the
223 	 * data.  If the data is truncated, or if we only want a partial
224 	 * coverage checksum (UDP-Lite), do it before the copy.
225 	 */
226 
227 	if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
228 		if (udp_lib_checksum_complete(skb))
229 			goto csum_copy_err;
230 	}
231 
232 	if (skb_csum_unnecessary(skb))
233 		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
234 					      msg->msg_iov, copied       );
235 	else {
236 		err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
237 		if (err == -EINVAL)
238 			goto csum_copy_err;
239 	}
240 	if (err)
241 		goto out_free;
242 
243 	if (!peeked) {
244 		if (is_udp4)
245 			UDP_INC_STATS_USER(sock_net(sk),
246 					UDP_MIB_INDATAGRAMS, is_udplite);
247 		else
248 			UDP6_INC_STATS_USER(sock_net(sk),
249 					UDP_MIB_INDATAGRAMS, is_udplite);
250 	}
251 
252 	sock_recv_timestamp(msg, sk, skb);
253 
254 	/* Copy the address. */
255 	if (msg->msg_name) {
256 		struct sockaddr_in6 *sin6;
257 
258 		sin6 = (struct sockaddr_in6 *) msg->msg_name;
259 		sin6->sin6_family = AF_INET6;
260 		sin6->sin6_port = udp_hdr(skb)->source;
261 		sin6->sin6_flowinfo = 0;
262 		sin6->sin6_scope_id = 0;
263 
264 		if (is_udp4)
265 			ipv6_addr_set(&sin6->sin6_addr, 0, 0,
266 				      htonl(0xffff), ip_hdr(skb)->saddr);
267 		else {
268 			ipv6_addr_copy(&sin6->sin6_addr,
269 				       &ipv6_hdr(skb)->saddr);
270 			if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
271 				sin6->sin6_scope_id = IP6CB(skb)->iif;
272 		}
273 
274 	}
275 	if (is_udp4) {
276 		if (inet->cmsg_flags)
277 			ip_cmsg_recv(msg, skb);
278 	} else {
279 		if (np->rxopt.all)
280 			datagram_recv_ctl(sk, msg, skb);
281 	}
282 
283 	err = copied;
284 	if (flags & MSG_TRUNC)
285 		err = ulen;
286 
287 out_free:
288 	lock_sock(sk);
289 	skb_free_datagram(sk, skb);
290 	release_sock(sk);
291 out:
292 	return err;
293 
294 csum_copy_err:
295 	lock_sock(sk);
296 	if (!skb_kill_datagram(sk, skb, flags)) {
297 		if (is_udp4)
298 			UDP_INC_STATS_USER(sock_net(sk),
299 					UDP_MIB_INERRORS, is_udplite);
300 		else
301 			UDP6_INC_STATS_USER(sock_net(sk),
302 					UDP_MIB_INERRORS, is_udplite);
303 	}
304 	release_sock(sk);
305 
306 	if (flags & MSG_DONTWAIT)
307 		return -EAGAIN;
308 	goto try_again;
309 }
310 
311 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
312 		    int type, int code, int offset, __be32 info,
313 		    struct udp_table *udptable)
314 {
315 	struct ipv6_pinfo *np;
316 	struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
317 	struct in6_addr *saddr = &hdr->saddr;
318 	struct in6_addr *daddr = &hdr->daddr;
319 	struct udphdr *uh = (struct udphdr*)(skb->data+offset);
320 	struct sock *sk;
321 	int err;
322 
323 	sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
324 			       saddr, uh->source, inet6_iif(skb), udptable);
325 	if (sk == NULL)
326 		return;
327 
328 	np = inet6_sk(sk);
329 
330 	if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
331 		goto out;
332 
333 	if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
334 		goto out;
335 
336 	if (np->recverr)
337 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
338 
339 	sk->sk_err = err;
340 	sk->sk_error_report(sk);
341 out:
342 	sock_put(sk);
343 }
344 
345 static __inline__ void udpv6_err(struct sk_buff *skb,
346 				 struct inet6_skb_parm *opt, int type,
347 				 int code, int offset, __be32 info     )
348 {
349 	__udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
350 }
351 
352 int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
353 {
354 	struct udp_sock *up = udp_sk(sk);
355 	int rc;
356 	int is_udplite = IS_UDPLITE(sk);
357 
358 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
359 		goto drop;
360 
361 	/*
362 	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
363 	 */
364 	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
365 
366 		if (up->pcrlen == 0) {          /* full coverage was set  */
367 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
368 				" %d while full coverage %d requested\n",
369 				UDP_SKB_CB(skb)->cscov, skb->len);
370 			goto drop;
371 		}
372 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
373 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
374 						    "too small, need min %d\n",
375 				       UDP_SKB_CB(skb)->cscov, up->pcrlen);
376 			goto drop;
377 		}
378 	}
379 
380 	if (sk->sk_filter) {
381 		if (udp_lib_checksum_complete(skb))
382 			goto drop;
383 	}
384 
385 	if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
386 		/* Note that an ENOMEM error is charged twice */
387 		if (rc == -ENOMEM) {
388 			UDP6_INC_STATS_BH(sock_net(sk),
389 					UDP_MIB_RCVBUFERRORS, is_udplite);
390 			atomic_inc(&sk->sk_drops);
391 		}
392 		goto drop;
393 	}
394 
395 	return 0;
396 drop:
397 	UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
398 	kfree_skb(skb);
399 	return -1;
400 }
401 
402 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
403 				      __be16 loc_port, struct in6_addr *loc_addr,
404 				      __be16 rmt_port, struct in6_addr *rmt_addr,
405 				      int dif)
406 {
407 	struct hlist_nulls_node *node;
408 	struct sock *s = sk;
409 	unsigned short num = ntohs(loc_port);
410 
411 	sk_nulls_for_each_from(s, node) {
412 		struct inet_sock *inet = inet_sk(s);
413 
414 		if (!net_eq(sock_net(s), net))
415 			continue;
416 
417 		if (s->sk_hash == num && s->sk_family == PF_INET6) {
418 			struct ipv6_pinfo *np = inet6_sk(s);
419 			if (inet->dport) {
420 				if (inet->dport != rmt_port)
421 					continue;
422 			}
423 			if (!ipv6_addr_any(&np->daddr) &&
424 			    !ipv6_addr_equal(&np->daddr, rmt_addr))
425 				continue;
426 
427 			if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
428 				continue;
429 
430 			if (!ipv6_addr_any(&np->rcv_saddr)) {
431 				if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
432 					continue;
433 			}
434 			if (!inet6_mc_check(s, loc_addr, rmt_addr))
435 				continue;
436 			return s;
437 		}
438 	}
439 	return NULL;
440 }
441 
442 /*
443  * Note: called only from the BH handler context,
444  * so we don't need to lock the hashes.
445  */
446 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
447 		struct in6_addr *saddr, struct in6_addr *daddr,
448 		struct udp_table *udptable)
449 {
450 	struct sock *sk, *sk2;
451 	const struct udphdr *uh = udp_hdr(skb);
452 	struct udp_hslot *hslot = &udptable->hash[udp_hashfn(net, ntohs(uh->dest))];
453 	int dif;
454 
455 	spin_lock(&hslot->lock);
456 	sk = sk_nulls_head(&hslot->head);
457 	dif = inet6_iif(skb);
458 	sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
459 	if (!sk) {
460 		kfree_skb(skb);
461 		goto out;
462 	}
463 
464 	sk2 = sk;
465 	while ((sk2 = udp_v6_mcast_next(net, sk_nulls_next(sk2), uh->dest, daddr,
466 					uh->source, saddr, dif))) {
467 		struct sk_buff *buff = skb_clone(skb, GFP_ATOMIC);
468 		if (buff) {
469 			bh_lock_sock(sk2);
470 			if (!sock_owned_by_user(sk2))
471 				udpv6_queue_rcv_skb(sk2, buff);
472 			else
473 				sk_add_backlog(sk2, buff);
474 			bh_unlock_sock(sk2);
475 		}
476 	}
477 	bh_lock_sock(sk);
478 	if (!sock_owned_by_user(sk))
479 		udpv6_queue_rcv_skb(sk, skb);
480 	else
481 		sk_add_backlog(sk, skb);
482 	bh_unlock_sock(sk);
483 out:
484 	spin_unlock(&hslot->lock);
485 	return 0;
486 }
487 
488 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
489 				 int proto)
490 {
491 	int err;
492 
493 	UDP_SKB_CB(skb)->partial_cov = 0;
494 	UDP_SKB_CB(skb)->cscov = skb->len;
495 
496 	if (proto == IPPROTO_UDPLITE) {
497 		err = udplite_checksum_init(skb, uh);
498 		if (err)
499 			return err;
500 	}
501 
502 	if (uh->check == 0) {
503 		/* RFC 2460 section 8.1 says that we SHOULD log
504 		   this error. Well, it is reasonable.
505 		 */
506 		LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
507 		return 1;
508 	}
509 	if (skb->ip_summed == CHECKSUM_COMPLETE &&
510 	    !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
511 			     skb->len, proto, skb->csum))
512 		skb->ip_summed = CHECKSUM_UNNECESSARY;
513 
514 	if (!skb_csum_unnecessary(skb))
515 		skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
516 							 &ipv6_hdr(skb)->daddr,
517 							 skb->len, proto, 0));
518 
519 	return 0;
520 }
521 
522 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
523 		   int proto)
524 {
525 	struct sock *sk;
526 	struct udphdr *uh;
527 	struct net_device *dev = skb->dev;
528 	struct in6_addr *saddr, *daddr;
529 	u32 ulen = 0;
530 	struct net *net = dev_net(skb->dev);
531 
532 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
533 		goto short_packet;
534 
535 	saddr = &ipv6_hdr(skb)->saddr;
536 	daddr = &ipv6_hdr(skb)->daddr;
537 	uh = udp_hdr(skb);
538 
539 	ulen = ntohs(uh->len);
540 	if (ulen > skb->len)
541 		goto short_packet;
542 
543 	if (proto == IPPROTO_UDP) {
544 		/* UDP validates ulen. */
545 
546 		/* Check for jumbo payload */
547 		if (ulen == 0)
548 			ulen = skb->len;
549 
550 		if (ulen < sizeof(*uh))
551 			goto short_packet;
552 
553 		if (ulen < skb->len) {
554 			if (pskb_trim_rcsum(skb, ulen))
555 				goto short_packet;
556 			saddr = &ipv6_hdr(skb)->saddr;
557 			daddr = &ipv6_hdr(skb)->daddr;
558 			uh = udp_hdr(skb);
559 		}
560 	}
561 
562 	if (udp6_csum_init(skb, uh, proto))
563 		goto discard;
564 
565 	/*
566 	 *	Multicast receive code
567 	 */
568 	if (ipv6_addr_is_multicast(daddr))
569 		return __udp6_lib_mcast_deliver(net, skb,
570 				saddr, daddr, udptable);
571 
572 	/* Unicast */
573 
574 	/*
575 	 * check socket cache ... must talk to Alan about his plans
576 	 * for sock caches... i'll skip this for now.
577 	 */
578 	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
579 
580 	if (sk == NULL) {
581 		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
582 			goto discard;
583 
584 		if (udp_lib_checksum_complete(skb))
585 			goto discard;
586 		UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS,
587 				proto == IPPROTO_UDPLITE);
588 
589 		icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0, dev);
590 
591 		kfree_skb(skb);
592 		return 0;
593 	}
594 
595 	/* deliver */
596 
597 	bh_lock_sock(sk);
598 	if (!sock_owned_by_user(sk))
599 		udpv6_queue_rcv_skb(sk, skb);
600 	else
601 		sk_add_backlog(sk, skb);
602 	bh_unlock_sock(sk);
603 	sock_put(sk);
604 	return 0;
605 
606 short_packet:
607 	LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: %d/%u\n",
608 		       proto == IPPROTO_UDPLITE ? "-Lite" : "",
609 		       ulen, skb->len);
610 
611 discard:
612 	UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
613 	kfree_skb(skb);
614 	return 0;
615 }
616 
617 static __inline__ int udpv6_rcv(struct sk_buff *skb)
618 {
619 	return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
620 }
621 
622 /*
623  * Throw away all pending data and cancel the corking. Socket is locked.
624  */
625 static void udp_v6_flush_pending_frames(struct sock *sk)
626 {
627 	struct udp_sock *up = udp_sk(sk);
628 
629 	if (up->pending == AF_INET)
630 		udp_flush_pending_frames(sk);
631 	else if (up->pending) {
632 		up->len = 0;
633 		up->pending = 0;
634 		ip6_flush_pending_frames(sk);
635 	}
636 }
637 
638 /*
639  *	Sending
640  */
641 
642 static int udp_v6_push_pending_frames(struct sock *sk)
643 {
644 	struct sk_buff *skb;
645 	struct udphdr *uh;
646 	struct udp_sock  *up = udp_sk(sk);
647 	struct inet_sock *inet = inet_sk(sk);
648 	struct flowi *fl = &inet->cork.fl;
649 	int err = 0;
650 	int is_udplite = IS_UDPLITE(sk);
651 	__wsum csum = 0;
652 
653 	/* Grab the skbuff where UDP header space exists. */
654 	if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
655 		goto out;
656 
657 	/*
658 	 * Create a UDP header
659 	 */
660 	uh = udp_hdr(skb);
661 	uh->source = fl->fl_ip_sport;
662 	uh->dest = fl->fl_ip_dport;
663 	uh->len = htons(up->len);
664 	uh->check = 0;
665 
666 	if (is_udplite)
667 		csum = udplite_csum_outgoing(sk, skb);
668 	 else
669 		csum = udp_csum_outgoing(sk, skb);
670 
671 	/* add protocol-dependent pseudo-header */
672 	uh->check = csum_ipv6_magic(&fl->fl6_src, &fl->fl6_dst,
673 				    up->len, fl->proto, csum   );
674 	if (uh->check == 0)
675 		uh->check = CSUM_MANGLED_0;
676 
677 	err = ip6_push_pending_frames(sk);
678 out:
679 	up->len = 0;
680 	up->pending = 0;
681 	if (!err)
682 		UDP6_INC_STATS_USER(sock_net(sk),
683 				UDP_MIB_OUTDATAGRAMS, is_udplite);
684 	return err;
685 }
686 
687 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
688 		  struct msghdr *msg, size_t len)
689 {
690 	struct ipv6_txoptions opt_space;
691 	struct udp_sock *up = udp_sk(sk);
692 	struct inet_sock *inet = inet_sk(sk);
693 	struct ipv6_pinfo *np = inet6_sk(sk);
694 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
695 	struct in6_addr *daddr, *final_p = NULL, final;
696 	struct ipv6_txoptions *opt = NULL;
697 	struct ip6_flowlabel *flowlabel = NULL;
698 	struct flowi fl;
699 	struct dst_entry *dst;
700 	int addr_len = msg->msg_namelen;
701 	int ulen = len;
702 	int hlimit = -1;
703 	int tclass = -1;
704 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
705 	int err;
706 	int connected = 0;
707 	int is_udplite = IS_UDPLITE(sk);
708 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
709 
710 	/* destination address check */
711 	if (sin6) {
712 		if (addr_len < offsetof(struct sockaddr, sa_data))
713 			return -EINVAL;
714 
715 		switch (sin6->sin6_family) {
716 		case AF_INET6:
717 			if (addr_len < SIN6_LEN_RFC2133)
718 				return -EINVAL;
719 			daddr = &sin6->sin6_addr;
720 			break;
721 		case AF_INET:
722 			goto do_udp_sendmsg;
723 		case AF_UNSPEC:
724 			msg->msg_name = sin6 = NULL;
725 			msg->msg_namelen = addr_len = 0;
726 			daddr = NULL;
727 			break;
728 		default:
729 			return -EINVAL;
730 		}
731 	} else if (!up->pending) {
732 		if (sk->sk_state != TCP_ESTABLISHED)
733 			return -EDESTADDRREQ;
734 		daddr = &np->daddr;
735 	} else
736 		daddr = NULL;
737 
738 	if (daddr) {
739 		if (ipv6_addr_v4mapped(daddr)) {
740 			struct sockaddr_in sin;
741 			sin.sin_family = AF_INET;
742 			sin.sin_port = sin6 ? sin6->sin6_port : inet->dport;
743 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
744 			msg->msg_name = &sin;
745 			msg->msg_namelen = sizeof(sin);
746 do_udp_sendmsg:
747 			if (__ipv6_only_sock(sk))
748 				return -ENETUNREACH;
749 			return udp_sendmsg(iocb, sk, msg, len);
750 		}
751 	}
752 
753 	if (up->pending == AF_INET)
754 		return udp_sendmsg(iocb, sk, msg, len);
755 
756 	/* Rough check on arithmetic overflow,
757 	   better check is made in ip6_append_data().
758 	   */
759 	if (len > INT_MAX - sizeof(struct udphdr))
760 		return -EMSGSIZE;
761 
762 	if (up->pending) {
763 		/*
764 		 * There are pending frames.
765 		 * The socket lock must be held while it's corked.
766 		 */
767 		lock_sock(sk);
768 		if (likely(up->pending)) {
769 			if (unlikely(up->pending != AF_INET6)) {
770 				release_sock(sk);
771 				return -EAFNOSUPPORT;
772 			}
773 			dst = NULL;
774 			goto do_append_data;
775 		}
776 		release_sock(sk);
777 	}
778 	ulen += sizeof(struct udphdr);
779 
780 	memset(&fl, 0, sizeof(fl));
781 
782 	if (sin6) {
783 		if (sin6->sin6_port == 0)
784 			return -EINVAL;
785 
786 		fl.fl_ip_dport = sin6->sin6_port;
787 		daddr = &sin6->sin6_addr;
788 
789 		if (np->sndflow) {
790 			fl.fl6_flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
791 			if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
792 				flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
793 				if (flowlabel == NULL)
794 					return -EINVAL;
795 				daddr = &flowlabel->dst;
796 			}
797 		}
798 
799 		/*
800 		 * Otherwise it will be difficult to maintain
801 		 * sk->sk_dst_cache.
802 		 */
803 		if (sk->sk_state == TCP_ESTABLISHED &&
804 		    ipv6_addr_equal(daddr, &np->daddr))
805 			daddr = &np->daddr;
806 
807 		if (addr_len >= sizeof(struct sockaddr_in6) &&
808 		    sin6->sin6_scope_id &&
809 		    ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
810 			fl.oif = sin6->sin6_scope_id;
811 	} else {
812 		if (sk->sk_state != TCP_ESTABLISHED)
813 			return -EDESTADDRREQ;
814 
815 		fl.fl_ip_dport = inet->dport;
816 		daddr = &np->daddr;
817 		fl.fl6_flowlabel = np->flow_label;
818 		connected = 1;
819 	}
820 
821 	if (!fl.oif)
822 		fl.oif = sk->sk_bound_dev_if;
823 
824 	if (!fl.oif)
825 		fl.oif = np->sticky_pktinfo.ipi6_ifindex;
826 
827 	if (msg->msg_controllen) {
828 		opt = &opt_space;
829 		memset(opt, 0, sizeof(struct ipv6_txoptions));
830 		opt->tot_len = sizeof(*opt);
831 
832 		err = datagram_send_ctl(sock_net(sk), msg, &fl, opt, &hlimit, &tclass);
833 		if (err < 0) {
834 			fl6_sock_release(flowlabel);
835 			return err;
836 		}
837 		if ((fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
838 			flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
839 			if (flowlabel == NULL)
840 				return -EINVAL;
841 		}
842 		if (!(opt->opt_nflen|opt->opt_flen))
843 			opt = NULL;
844 		connected = 0;
845 	}
846 	if (opt == NULL)
847 		opt = np->opt;
848 	if (flowlabel)
849 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
850 	opt = ipv6_fixup_options(&opt_space, opt);
851 
852 	fl.proto = sk->sk_protocol;
853 	if (!ipv6_addr_any(daddr))
854 		ipv6_addr_copy(&fl.fl6_dst, daddr);
855 	else
856 		fl.fl6_dst.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
857 	if (ipv6_addr_any(&fl.fl6_src) && !ipv6_addr_any(&np->saddr))
858 		ipv6_addr_copy(&fl.fl6_src, &np->saddr);
859 	fl.fl_ip_sport = inet->sport;
860 
861 	/* merge ip6_build_xmit from ip6_output */
862 	if (opt && opt->srcrt) {
863 		struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
864 		ipv6_addr_copy(&final, &fl.fl6_dst);
865 		ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
866 		final_p = &final;
867 		connected = 0;
868 	}
869 
870 	if (!fl.oif && ipv6_addr_is_multicast(&fl.fl6_dst)) {
871 		fl.oif = np->mcast_oif;
872 		connected = 0;
873 	}
874 
875 	security_sk_classify_flow(sk, &fl);
876 
877 	err = ip6_sk_dst_lookup(sk, &dst, &fl);
878 	if (err)
879 		goto out;
880 	if (final_p)
881 		ipv6_addr_copy(&fl.fl6_dst, final_p);
882 
883 	err = __xfrm_lookup(sock_net(sk), &dst, &fl, sk, XFRM_LOOKUP_WAIT);
884 	if (err < 0) {
885 		if (err == -EREMOTE)
886 			err = ip6_dst_blackhole(sk, &dst, &fl);
887 		if (err < 0)
888 			goto out;
889 	}
890 
891 	if (hlimit < 0) {
892 		if (ipv6_addr_is_multicast(&fl.fl6_dst))
893 			hlimit = np->mcast_hops;
894 		else
895 			hlimit = np->hop_limit;
896 		if (hlimit < 0)
897 			hlimit = ip6_dst_hoplimit(dst);
898 	}
899 
900 	if (tclass < 0) {
901 		tclass = np->tclass;
902 		if (tclass < 0)
903 			tclass = 0;
904 	}
905 
906 	if (msg->msg_flags&MSG_CONFIRM)
907 		goto do_confirm;
908 back_from_confirm:
909 
910 	lock_sock(sk);
911 	if (unlikely(up->pending)) {
912 		/* The socket is already corked while preparing it. */
913 		/* ... which is an evident application bug. --ANK */
914 		release_sock(sk);
915 
916 		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
917 		err = -EINVAL;
918 		goto out;
919 	}
920 
921 	up->pending = AF_INET6;
922 
923 do_append_data:
924 	up->len += ulen;
925 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
926 	err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
927 		sizeof(struct udphdr), hlimit, tclass, opt, &fl,
928 		(struct rt6_info*)dst,
929 		corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
930 	if (err)
931 		udp_v6_flush_pending_frames(sk);
932 	else if (!corkreq)
933 		err = udp_v6_push_pending_frames(sk);
934 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
935 		up->pending = 0;
936 
937 	if (dst) {
938 		if (connected) {
939 			ip6_dst_store(sk, dst,
940 				      ipv6_addr_equal(&fl.fl6_dst, &np->daddr) ?
941 				      &np->daddr : NULL,
942 #ifdef CONFIG_IPV6_SUBTREES
943 				      ipv6_addr_equal(&fl.fl6_src, &np->saddr) ?
944 				      &np->saddr :
945 #endif
946 				      NULL);
947 		} else {
948 			dst_release(dst);
949 		}
950 		dst = NULL;
951 	}
952 
953 	if (err > 0)
954 		err = np->recverr ? net_xmit_errno(err) : 0;
955 	release_sock(sk);
956 out:
957 	dst_release(dst);
958 	fl6_sock_release(flowlabel);
959 	if (!err)
960 		return len;
961 	/*
962 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
963 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
964 	 * we don't have a good statistic (IpOutDiscards but it can be too many
965 	 * things).  We could add another new stat but at least for now that
966 	 * seems like overkill.
967 	 */
968 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
969 		UDP6_INC_STATS_USER(sock_net(sk),
970 				UDP_MIB_SNDBUFERRORS, is_udplite);
971 	}
972 	return err;
973 
974 do_confirm:
975 	dst_confirm(dst);
976 	if (!(msg->msg_flags&MSG_PROBE) || len)
977 		goto back_from_confirm;
978 	err = 0;
979 	goto out;
980 }
981 
982 void udpv6_destroy_sock(struct sock *sk)
983 {
984 	lock_sock(sk);
985 	udp_v6_flush_pending_frames(sk);
986 	release_sock(sk);
987 
988 	inet6_destroy_sock(sk);
989 }
990 
991 /*
992  *	Socket option code for UDP
993  */
994 int udpv6_setsockopt(struct sock *sk, int level, int optname,
995 		     char __user *optval, int optlen)
996 {
997 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
998 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
999 					  udp_v6_push_pending_frames);
1000 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
1001 }
1002 
1003 #ifdef CONFIG_COMPAT
1004 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1005 			    char __user *optval, int optlen)
1006 {
1007 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1008 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1009 					  udp_v6_push_pending_frames);
1010 	return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1011 }
1012 #endif
1013 
1014 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1015 		     char __user *optval, int __user *optlen)
1016 {
1017 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1018 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1019 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
1020 }
1021 
1022 #ifdef CONFIG_COMPAT
1023 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1024 			    char __user *optval, int __user *optlen)
1025 {
1026 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1027 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1028 	return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1029 }
1030 #endif
1031 
1032 static struct inet6_protocol udpv6_protocol = {
1033 	.handler	=	udpv6_rcv,
1034 	.err_handler	=	udpv6_err,
1035 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1036 };
1037 
1038 /* ------------------------------------------------------------------------ */
1039 #ifdef CONFIG_PROC_FS
1040 
1041 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
1042 {
1043 	struct inet_sock *inet = inet_sk(sp);
1044 	struct ipv6_pinfo *np = inet6_sk(sp);
1045 	struct in6_addr *dest, *src;
1046 	__u16 destp, srcp;
1047 
1048 	dest  = &np->daddr;
1049 	src   = &np->rcv_saddr;
1050 	destp = ntohs(inet->dport);
1051 	srcp  = ntohs(inet->sport);
1052 	seq_printf(seq,
1053 		   "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1054 		   "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d\n",
1055 		   bucket,
1056 		   src->s6_addr32[0], src->s6_addr32[1],
1057 		   src->s6_addr32[2], src->s6_addr32[3], srcp,
1058 		   dest->s6_addr32[0], dest->s6_addr32[1],
1059 		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
1060 		   sp->sk_state,
1061 		   atomic_read(&sp->sk_wmem_alloc),
1062 		   atomic_read(&sp->sk_rmem_alloc),
1063 		   0, 0L, 0,
1064 		   sock_i_uid(sp), 0,
1065 		   sock_i_ino(sp),
1066 		   atomic_read(&sp->sk_refcnt), sp,
1067 		   atomic_read(&sp->sk_drops));
1068 }
1069 
1070 int udp6_seq_show(struct seq_file *seq, void *v)
1071 {
1072 	if (v == SEQ_START_TOKEN)
1073 		seq_printf(seq,
1074 			   "  sl  "
1075 			   "local_address                         "
1076 			   "remote_address                        "
1077 			   "st tx_queue rx_queue tr tm->when retrnsmt"
1078 			   "   uid  timeout inode ref pointer drops\n");
1079 	else
1080 		udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1081 	return 0;
1082 }
1083 
1084 static struct udp_seq_afinfo udp6_seq_afinfo = {
1085 	.name		= "udp6",
1086 	.family		= AF_INET6,
1087 	.udp_table	= &udp_table,
1088 	.seq_fops	= {
1089 		.owner	=	THIS_MODULE,
1090 	},
1091 	.seq_ops	= {
1092 		.show		= udp6_seq_show,
1093 	},
1094 };
1095 
1096 int udp6_proc_init(struct net *net)
1097 {
1098 	return udp_proc_register(net, &udp6_seq_afinfo);
1099 }
1100 
1101 void udp6_proc_exit(struct net *net) {
1102 	udp_proc_unregister(net, &udp6_seq_afinfo);
1103 }
1104 #endif /* CONFIG_PROC_FS */
1105 
1106 /* ------------------------------------------------------------------------ */
1107 
1108 struct proto udpv6_prot = {
1109 	.name		   = "UDPv6",
1110 	.owner		   = THIS_MODULE,
1111 	.close		   = udp_lib_close,
1112 	.connect	   = ip6_datagram_connect,
1113 	.disconnect	   = udp_disconnect,
1114 	.ioctl		   = udp_ioctl,
1115 	.destroy	   = udpv6_destroy_sock,
1116 	.setsockopt	   = udpv6_setsockopt,
1117 	.getsockopt	   = udpv6_getsockopt,
1118 	.sendmsg	   = udpv6_sendmsg,
1119 	.recvmsg	   = udpv6_recvmsg,
1120 	.backlog_rcv	   = udpv6_queue_rcv_skb,
1121 	.hash		   = udp_lib_hash,
1122 	.unhash		   = udp_lib_unhash,
1123 	.get_port	   = udp_v6_get_port,
1124 	.memory_allocated  = &udp_memory_allocated,
1125 	.sysctl_mem	   = sysctl_udp_mem,
1126 	.sysctl_wmem	   = &sysctl_udp_wmem_min,
1127 	.sysctl_rmem	   = &sysctl_udp_rmem_min,
1128 	.obj_size	   = sizeof(struct udp6_sock),
1129 	.slab_flags	   = SLAB_DESTROY_BY_RCU,
1130 	.h.udp_table	   = &udp_table,
1131 #ifdef CONFIG_COMPAT
1132 	.compat_setsockopt = compat_udpv6_setsockopt,
1133 	.compat_getsockopt = compat_udpv6_getsockopt,
1134 #endif
1135 };
1136 
1137 static struct inet_protosw udpv6_protosw = {
1138 	.type =      SOCK_DGRAM,
1139 	.protocol =  IPPROTO_UDP,
1140 	.prot =      &udpv6_prot,
1141 	.ops =       &inet6_dgram_ops,
1142 	.capability =-1,
1143 	.no_check =  UDP_CSUM_DEFAULT,
1144 	.flags =     INET_PROTOSW_PERMANENT,
1145 };
1146 
1147 
1148 int __init udpv6_init(void)
1149 {
1150 	int ret;
1151 
1152 	ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1153 	if (ret)
1154 		goto out;
1155 
1156 	ret = inet6_register_protosw(&udpv6_protosw);
1157 	if (ret)
1158 		goto out_udpv6_protocol;
1159 out:
1160 	return ret;
1161 
1162 out_udpv6_protocol:
1163 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1164 	goto out;
1165 }
1166 
1167 void udpv6_exit(void)
1168 {
1169 	inet6_unregister_protosw(&udpv6_protosw);
1170 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1171 }
1172