xref: /linux/net/ipv6/udp.c (revision e5c5d22e8dcf7c2d430336cbf8e180bd38e8daf1)
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 <linux/slab.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 #include <net/inet6_hashtables.h>
49 
50 #include <linux/proc_fs.h>
51 #include <linux/seq_file.h>
52 #include <trace/events/skb.h>
53 #include "udp_impl.h"
54 
55 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
56 {
57 	const struct in6_addr *sk_rcv_saddr6 = &inet6_sk(sk)->rcv_saddr;
58 	const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
59 	__be32 sk1_rcv_saddr = sk_rcv_saddr(sk);
60 	__be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
61 	int sk_ipv6only = ipv6_only_sock(sk);
62 	int sk2_ipv6only = inet_v6_ipv6only(sk2);
63 	int addr_type = ipv6_addr_type(sk_rcv_saddr6);
64 	int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
65 
66 	/* if both are mapped, treat as IPv4 */
67 	if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
68 		return (!sk2_ipv6only &&
69 			(!sk1_rcv_saddr || !sk2_rcv_saddr ||
70 			  sk1_rcv_saddr == sk2_rcv_saddr));
71 
72 	if (addr_type2 == IPV6_ADDR_ANY &&
73 	    !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
74 		return 1;
75 
76 	if (addr_type == IPV6_ADDR_ANY &&
77 	    !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
78 		return 1;
79 
80 	if (sk2_rcv_saddr6 &&
81 	    ipv6_addr_equal(sk_rcv_saddr6, sk2_rcv_saddr6))
82 		return 1;
83 
84 	return 0;
85 }
86 
87 static unsigned int udp6_portaddr_hash(struct net *net,
88 				       const struct in6_addr *addr6,
89 				       unsigned int port)
90 {
91 	unsigned int hash, mix = net_hash_mix(net);
92 
93 	if (ipv6_addr_any(addr6))
94 		hash = jhash_1word(0, mix);
95 	else if (ipv6_addr_v4mapped(addr6))
96 		hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
97 	else
98 		hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
99 
100 	return hash ^ port;
101 }
102 
103 
104 int udp_v6_get_port(struct sock *sk, unsigned short snum)
105 {
106 	unsigned int hash2_nulladdr =
107 		udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
108 	unsigned int hash2_partial =
109 		udp6_portaddr_hash(sock_net(sk), &inet6_sk(sk)->rcv_saddr, 0);
110 
111 	/* precompute partial secondary hash */
112 	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
113 	return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
114 }
115 
116 static void udp_v6_rehash(struct sock *sk)
117 {
118 	u16 new_hash = udp6_portaddr_hash(sock_net(sk),
119 					  &inet6_sk(sk)->rcv_saddr,
120 					  inet_sk(sk)->inet_num);
121 
122 	udp_lib_rehash(sk, new_hash);
123 }
124 
125 static inline int compute_score(struct sock *sk, struct net *net,
126 				unsigned short hnum,
127 				const struct in6_addr *saddr, __be16 sport,
128 				const struct in6_addr *daddr, __be16 dport,
129 				int dif)
130 {
131 	int score = -1;
132 
133 	if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
134 			sk->sk_family == PF_INET6) {
135 		struct ipv6_pinfo *np = inet6_sk(sk);
136 		struct inet_sock *inet = inet_sk(sk);
137 
138 		score = 0;
139 		if (inet->inet_dport) {
140 			if (inet->inet_dport != sport)
141 				return -1;
142 			score++;
143 		}
144 		if (!ipv6_addr_any(&np->rcv_saddr)) {
145 			if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
146 				return -1;
147 			score++;
148 		}
149 		if (!ipv6_addr_any(&np->daddr)) {
150 			if (!ipv6_addr_equal(&np->daddr, saddr))
151 				return -1;
152 			score++;
153 		}
154 		if (sk->sk_bound_dev_if) {
155 			if (sk->sk_bound_dev_if != dif)
156 				return -1;
157 			score++;
158 		}
159 	}
160 	return score;
161 }
162 
163 #define SCORE2_MAX (1 + 1 + 1)
164 static inline int compute_score2(struct sock *sk, struct net *net,
165 				const struct in6_addr *saddr, __be16 sport,
166 				const struct in6_addr *daddr, unsigned short hnum,
167 				int dif)
168 {
169 	int score = -1;
170 
171 	if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
172 			sk->sk_family == PF_INET6) {
173 		struct ipv6_pinfo *np = inet6_sk(sk);
174 		struct inet_sock *inet = inet_sk(sk);
175 
176 		if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
177 			return -1;
178 		score = 0;
179 		if (inet->inet_dport) {
180 			if (inet->inet_dport != sport)
181 				return -1;
182 			score++;
183 		}
184 		if (!ipv6_addr_any(&np->daddr)) {
185 			if (!ipv6_addr_equal(&np->daddr, saddr))
186 				return -1;
187 			score++;
188 		}
189 		if (sk->sk_bound_dev_if) {
190 			if (sk->sk_bound_dev_if != dif)
191 				return -1;
192 			score++;
193 		}
194 	}
195 	return score;
196 }
197 
198 
199 /* called with read_rcu_lock() */
200 static struct sock *udp6_lib_lookup2(struct net *net,
201 		const struct in6_addr *saddr, __be16 sport,
202 		const struct in6_addr *daddr, unsigned int hnum, int dif,
203 		struct udp_hslot *hslot2, unsigned int slot2)
204 {
205 	struct sock *sk, *result;
206 	struct hlist_nulls_node *node;
207 	int score, badness, matches = 0, reuseport = 0;
208 	u32 hash = 0;
209 
210 begin:
211 	result = NULL;
212 	badness = -1;
213 	udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
214 		score = compute_score2(sk, net, saddr, sport,
215 				      daddr, hnum, dif);
216 		if (score > badness) {
217 			result = sk;
218 			badness = score;
219 			reuseport = sk->sk_reuseport;
220 			if (reuseport) {
221 				hash = inet6_ehashfn(net, daddr, hnum,
222 						     saddr, sport);
223 				matches = 1;
224 			} else if (score == SCORE2_MAX)
225 				goto exact_match;
226 		} else if (score == badness && reuseport) {
227 			matches++;
228 			if (((u64)hash * matches) >> 32 == 0)
229 				result = sk;
230 			hash = next_pseudo_random32(hash);
231 		}
232 	}
233 	/*
234 	 * if the nulls value we got at the end of this lookup is
235 	 * not the expected one, we must restart lookup.
236 	 * We probably met an item that was moved to another chain.
237 	 */
238 	if (get_nulls_value(node) != slot2)
239 		goto begin;
240 
241 	if (result) {
242 exact_match:
243 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
244 			result = NULL;
245 		else if (unlikely(compute_score2(result, net, saddr, sport,
246 				  daddr, hnum, dif) < badness)) {
247 			sock_put(result);
248 			goto begin;
249 		}
250 	}
251 	return result;
252 }
253 
254 struct sock *__udp6_lib_lookup(struct net *net,
255 				      const struct in6_addr *saddr, __be16 sport,
256 				      const struct in6_addr *daddr, __be16 dport,
257 				      int dif, struct udp_table *udptable)
258 {
259 	struct sock *sk, *result;
260 	struct hlist_nulls_node *node;
261 	unsigned short hnum = ntohs(dport);
262 	unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
263 	struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
264 	int score, badness, matches = 0, reuseport = 0;
265 	u32 hash = 0;
266 
267 	rcu_read_lock();
268 	if (hslot->count > 10) {
269 		hash2 = udp6_portaddr_hash(net, daddr, hnum);
270 		slot2 = hash2 & udptable->mask;
271 		hslot2 = &udptable->hash2[slot2];
272 		if (hslot->count < hslot2->count)
273 			goto begin;
274 
275 		result = udp6_lib_lookup2(net, saddr, sport,
276 					  daddr, hnum, dif,
277 					  hslot2, slot2);
278 		if (!result) {
279 			hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
280 			slot2 = hash2 & udptable->mask;
281 			hslot2 = &udptable->hash2[slot2];
282 			if (hslot->count < hslot2->count)
283 				goto begin;
284 
285 			result = udp6_lib_lookup2(net, saddr, sport,
286 						  &in6addr_any, hnum, dif,
287 						  hslot2, slot2);
288 		}
289 		rcu_read_unlock();
290 		return result;
291 	}
292 begin:
293 	result = NULL;
294 	badness = -1;
295 	sk_nulls_for_each_rcu(sk, node, &hslot->head) {
296 		score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
297 		if (score > badness) {
298 			result = sk;
299 			badness = score;
300 			reuseport = sk->sk_reuseport;
301 			if (reuseport) {
302 				hash = inet6_ehashfn(net, daddr, hnum,
303 						     saddr, sport);
304 				matches = 1;
305 			}
306 		} else if (score == badness && reuseport) {
307 			matches++;
308 			if (((u64)hash * matches) >> 32 == 0)
309 				result = sk;
310 			hash = next_pseudo_random32(hash);
311 		}
312 	}
313 	/*
314 	 * if the nulls value we got at the end of this lookup is
315 	 * not the expected one, we must restart lookup.
316 	 * We probably met an item that was moved to another chain.
317 	 */
318 	if (get_nulls_value(node) != slot)
319 		goto begin;
320 
321 	if (result) {
322 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
323 			result = NULL;
324 		else if (unlikely(compute_score(result, net, hnum, saddr, sport,
325 					daddr, dport, dif) < badness)) {
326 			sock_put(result);
327 			goto begin;
328 		}
329 	}
330 	rcu_read_unlock();
331 	return result;
332 }
333 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
334 
335 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
336 					  __be16 sport, __be16 dport,
337 					  struct udp_table *udptable)
338 {
339 	struct sock *sk;
340 	const struct ipv6hdr *iph = ipv6_hdr(skb);
341 
342 	if (unlikely(sk = skb_steal_sock(skb)))
343 		return sk;
344 	return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
345 				 &iph->daddr, dport, inet6_iif(skb),
346 				 udptable);
347 }
348 
349 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
350 			     const struct in6_addr *daddr, __be16 dport, int dif)
351 {
352 	return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
353 }
354 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
355 
356 
357 /*
358  * 	This should be easy, if there is something there we
359  * 	return it, otherwise we block.
360  */
361 
362 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
363 		  struct msghdr *msg, size_t len,
364 		  int noblock, int flags, int *addr_len)
365 {
366 	struct ipv6_pinfo *np = inet6_sk(sk);
367 	struct inet_sock *inet = inet_sk(sk);
368 	struct sk_buff *skb;
369 	unsigned int ulen, copied;
370 	int peeked, off = 0;
371 	int err;
372 	int is_udplite = IS_UDPLITE(sk);
373 	int is_udp4;
374 	bool slow;
375 
376 	if (addr_len)
377 		*addr_len = sizeof(struct sockaddr_in6);
378 
379 	if (flags & MSG_ERRQUEUE)
380 		return ipv6_recv_error(sk, msg, len);
381 
382 	if (np->rxpmtu && np->rxopt.bits.rxpmtu)
383 		return ipv6_recv_rxpmtu(sk, msg, len);
384 
385 try_again:
386 	skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
387 				  &peeked, &off, &err);
388 	if (!skb)
389 		goto out;
390 
391 	ulen = skb->len - sizeof(struct udphdr);
392 	copied = len;
393 	if (copied > ulen)
394 		copied = ulen;
395 	else if (copied < ulen)
396 		msg->msg_flags |= MSG_TRUNC;
397 
398 	is_udp4 = (skb->protocol == htons(ETH_P_IP));
399 
400 	/*
401 	 * If checksum is needed at all, try to do it while copying the
402 	 * data.  If the data is truncated, or if we only want a partial
403 	 * coverage checksum (UDP-Lite), do it before the copy.
404 	 */
405 
406 	if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
407 		if (udp_lib_checksum_complete(skb))
408 			goto csum_copy_err;
409 	}
410 
411 	if (skb_csum_unnecessary(skb))
412 		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
413 					      msg->msg_iov, copied);
414 	else {
415 		err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
416 		if (err == -EINVAL)
417 			goto csum_copy_err;
418 	}
419 	if (unlikely(err)) {
420 		trace_kfree_skb(skb, udpv6_recvmsg);
421 		if (!peeked) {
422 			atomic_inc(&sk->sk_drops);
423 			if (is_udp4)
424 				UDP_INC_STATS_USER(sock_net(sk),
425 						   UDP_MIB_INERRORS,
426 						   is_udplite);
427 			else
428 				UDP6_INC_STATS_USER(sock_net(sk),
429 						    UDP_MIB_INERRORS,
430 						    is_udplite);
431 		}
432 		goto out_free;
433 	}
434 	if (!peeked) {
435 		if (is_udp4)
436 			UDP_INC_STATS_USER(sock_net(sk),
437 					UDP_MIB_INDATAGRAMS, is_udplite);
438 		else
439 			UDP6_INC_STATS_USER(sock_net(sk),
440 					UDP_MIB_INDATAGRAMS, is_udplite);
441 	}
442 
443 	sock_recv_ts_and_drops(msg, sk, skb);
444 
445 	/* Copy the address. */
446 	if (msg->msg_name) {
447 		struct sockaddr_in6 *sin6;
448 
449 		sin6 = (struct sockaddr_in6 *) msg->msg_name;
450 		sin6->sin6_family = AF_INET6;
451 		sin6->sin6_port = udp_hdr(skb)->source;
452 		sin6->sin6_flowinfo = 0;
453 
454 		if (is_udp4) {
455 			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
456 					       &sin6->sin6_addr);
457 			sin6->sin6_scope_id = 0;
458 		} else {
459 			sin6->sin6_addr = ipv6_hdr(skb)->saddr;
460 			sin6->sin6_scope_id =
461 				ipv6_iface_scope_id(&sin6->sin6_addr,
462 						    IP6CB(skb)->iif);
463 		}
464 
465 	}
466 	if (is_udp4) {
467 		if (inet->cmsg_flags)
468 			ip_cmsg_recv(msg, skb);
469 	} else {
470 		if (np->rxopt.all)
471 			ip6_datagram_recv_ctl(sk, msg, skb);
472 	}
473 
474 	err = copied;
475 	if (flags & MSG_TRUNC)
476 		err = ulen;
477 
478 out_free:
479 	skb_free_datagram_locked(sk, skb);
480 out:
481 	return err;
482 
483 csum_copy_err:
484 	slow = lock_sock_fast(sk);
485 	if (!skb_kill_datagram(sk, skb, flags)) {
486 		if (is_udp4)
487 			UDP_INC_STATS_USER(sock_net(sk),
488 					UDP_MIB_INERRORS, is_udplite);
489 		else
490 			UDP6_INC_STATS_USER(sock_net(sk),
491 					UDP_MIB_INERRORS, is_udplite);
492 	}
493 	unlock_sock_fast(sk, slow);
494 
495 	if (noblock)
496 		return -EAGAIN;
497 
498 	/* starting over for a new packet */
499 	msg->msg_flags &= ~MSG_TRUNC;
500 	goto try_again;
501 }
502 
503 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
504 		    u8 type, u8 code, int offset, __be32 info,
505 		    struct udp_table *udptable)
506 {
507 	struct ipv6_pinfo *np;
508 	const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
509 	const struct in6_addr *saddr = &hdr->saddr;
510 	const struct in6_addr *daddr = &hdr->daddr;
511 	struct udphdr *uh = (struct udphdr*)(skb->data+offset);
512 	struct sock *sk;
513 	int err;
514 
515 	sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
516 			       saddr, uh->source, inet6_iif(skb), udptable);
517 	if (sk == NULL)
518 		return;
519 
520 	if (type == ICMPV6_PKT_TOOBIG)
521 		ip6_sk_update_pmtu(skb, sk, info);
522 	if (type == NDISC_REDIRECT)
523 		ip6_sk_redirect(skb, sk);
524 
525 	np = inet6_sk(sk);
526 
527 	if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
528 		goto out;
529 
530 	if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
531 		goto out;
532 
533 	if (np->recverr)
534 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
535 
536 	sk->sk_err = err;
537 	sk->sk_error_report(sk);
538 out:
539 	sock_put(sk);
540 }
541 
542 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
543 {
544 	int rc;
545 
546 	if (!ipv6_addr_any(&inet6_sk(sk)->daddr))
547 		sock_rps_save_rxhash(sk, skb);
548 
549 	rc = sock_queue_rcv_skb(sk, skb);
550 	if (rc < 0) {
551 		int is_udplite = IS_UDPLITE(sk);
552 
553 		/* Note that an ENOMEM error is charged twice */
554 		if (rc == -ENOMEM)
555 			UDP6_INC_STATS_BH(sock_net(sk),
556 					UDP_MIB_RCVBUFERRORS, is_udplite);
557 		UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
558 		kfree_skb(skb);
559 		return -1;
560 	}
561 	return 0;
562 }
563 
564 static __inline__ void udpv6_err(struct sk_buff *skb,
565 				 struct inet6_skb_parm *opt, u8 type,
566 				 u8 code, int offset, __be32 info     )
567 {
568 	__udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
569 }
570 
571 static struct static_key udpv6_encap_needed __read_mostly;
572 void udpv6_encap_enable(void)
573 {
574 	if (!static_key_enabled(&udpv6_encap_needed))
575 		static_key_slow_inc(&udpv6_encap_needed);
576 }
577 EXPORT_SYMBOL(udpv6_encap_enable);
578 
579 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
580 {
581 	struct udp_sock *up = udp_sk(sk);
582 	int rc;
583 	int is_udplite = IS_UDPLITE(sk);
584 
585 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
586 		goto drop;
587 
588 	if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
589 		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
590 
591 		/*
592 		 * This is an encapsulation socket so pass the skb to
593 		 * the socket's udp_encap_rcv() hook. Otherwise, just
594 		 * fall through and pass this up the UDP socket.
595 		 * up->encap_rcv() returns the following value:
596 		 * =0 if skb was successfully passed to the encap
597 		 *    handler or was discarded by it.
598 		 * >0 if skb should be passed on to UDP.
599 		 * <0 if skb should be resubmitted as proto -N
600 		 */
601 
602 		/* if we're overly short, let UDP handle it */
603 		encap_rcv = ACCESS_ONCE(up->encap_rcv);
604 		if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
605 			int ret;
606 
607 			ret = encap_rcv(sk, skb);
608 			if (ret <= 0) {
609 				UDP_INC_STATS_BH(sock_net(sk),
610 						 UDP_MIB_INDATAGRAMS,
611 						 is_udplite);
612 				return -ret;
613 			}
614 		}
615 
616 		/* FALLTHROUGH -- it's a UDP Packet */
617 	}
618 
619 	/*
620 	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
621 	 */
622 	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
623 
624 		if (up->pcrlen == 0) {          /* full coverage was set  */
625 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
626 				" %d while full coverage %d requested\n",
627 				UDP_SKB_CB(skb)->cscov, skb->len);
628 			goto drop;
629 		}
630 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
631 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
632 						    "too small, need min %d\n",
633 				       UDP_SKB_CB(skb)->cscov, up->pcrlen);
634 			goto drop;
635 		}
636 	}
637 
638 	if (rcu_access_pointer(sk->sk_filter)) {
639 		if (udp_lib_checksum_complete(skb))
640 			goto drop;
641 	}
642 
643 	if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
644 		goto drop;
645 
646 	skb_dst_drop(skb);
647 
648 	bh_lock_sock(sk);
649 	rc = 0;
650 	if (!sock_owned_by_user(sk))
651 		rc = __udpv6_queue_rcv_skb(sk, skb);
652 	else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
653 		bh_unlock_sock(sk);
654 		goto drop;
655 	}
656 	bh_unlock_sock(sk);
657 
658 	return rc;
659 drop:
660 	UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
661 	atomic_inc(&sk->sk_drops);
662 	kfree_skb(skb);
663 	return -1;
664 }
665 
666 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
667 				      __be16 loc_port, const struct in6_addr *loc_addr,
668 				      __be16 rmt_port, const struct in6_addr *rmt_addr,
669 				      int dif)
670 {
671 	struct hlist_nulls_node *node;
672 	struct sock *s = sk;
673 	unsigned short num = ntohs(loc_port);
674 
675 	sk_nulls_for_each_from(s, node) {
676 		struct inet_sock *inet = inet_sk(s);
677 
678 		if (!net_eq(sock_net(s), net))
679 			continue;
680 
681 		if (udp_sk(s)->udp_port_hash == num &&
682 		    s->sk_family == PF_INET6) {
683 			struct ipv6_pinfo *np = inet6_sk(s);
684 			if (inet->inet_dport) {
685 				if (inet->inet_dport != rmt_port)
686 					continue;
687 			}
688 			if (!ipv6_addr_any(&np->daddr) &&
689 			    !ipv6_addr_equal(&np->daddr, rmt_addr))
690 				continue;
691 
692 			if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
693 				continue;
694 
695 			if (!ipv6_addr_any(&np->rcv_saddr)) {
696 				if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
697 					continue;
698 			}
699 			if (!inet6_mc_check(s, loc_addr, rmt_addr))
700 				continue;
701 			return s;
702 		}
703 	}
704 	return NULL;
705 }
706 
707 static void flush_stack(struct sock **stack, unsigned int count,
708 			struct sk_buff *skb, unsigned int final)
709 {
710 	struct sk_buff *skb1 = NULL;
711 	struct sock *sk;
712 	unsigned int i;
713 
714 	for (i = 0; i < count; i++) {
715 		sk = stack[i];
716 		if (likely(skb1 == NULL))
717 			skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
718 		if (!skb1) {
719 			atomic_inc(&sk->sk_drops);
720 			UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
721 					  IS_UDPLITE(sk));
722 			UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
723 					  IS_UDPLITE(sk));
724 		}
725 
726 		if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
727 			skb1 = NULL;
728 	}
729 	if (unlikely(skb1))
730 		kfree_skb(skb1);
731 }
732 /*
733  * Note: called only from the BH handler context,
734  * so we don't need to lock the hashes.
735  */
736 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
737 		const struct in6_addr *saddr, const struct in6_addr *daddr,
738 		struct udp_table *udptable)
739 {
740 	struct sock *sk, *stack[256 / sizeof(struct sock *)];
741 	const struct udphdr *uh = udp_hdr(skb);
742 	struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
743 	int dif;
744 	unsigned int i, count = 0;
745 
746 	spin_lock(&hslot->lock);
747 	sk = sk_nulls_head(&hslot->head);
748 	dif = inet6_iif(skb);
749 	sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
750 	while (sk) {
751 		stack[count++] = sk;
752 		sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
753 				       uh->source, saddr, dif);
754 		if (unlikely(count == ARRAY_SIZE(stack))) {
755 			if (!sk)
756 				break;
757 			flush_stack(stack, count, skb, ~0);
758 			count = 0;
759 		}
760 	}
761 	/*
762 	 * before releasing the lock, we must take reference on sockets
763 	 */
764 	for (i = 0; i < count; i++)
765 		sock_hold(stack[i]);
766 
767 	spin_unlock(&hslot->lock);
768 
769 	if (count) {
770 		flush_stack(stack, count, skb, count - 1);
771 
772 		for (i = 0; i < count; i++)
773 			sock_put(stack[i]);
774 	} else {
775 		kfree_skb(skb);
776 	}
777 	return 0;
778 }
779 
780 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
781 		   int proto)
782 {
783 	struct net *net = dev_net(skb->dev);
784 	struct sock *sk;
785 	struct udphdr *uh;
786 	const struct in6_addr *saddr, *daddr;
787 	u32 ulen = 0;
788 
789 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
790 		goto discard;
791 
792 	saddr = &ipv6_hdr(skb)->saddr;
793 	daddr = &ipv6_hdr(skb)->daddr;
794 	uh = udp_hdr(skb);
795 
796 	ulen = ntohs(uh->len);
797 	if (ulen > skb->len)
798 		goto short_packet;
799 
800 	if (proto == IPPROTO_UDP) {
801 		/* UDP validates ulen. */
802 
803 		/* Check for jumbo payload */
804 		if (ulen == 0)
805 			ulen = skb->len;
806 
807 		if (ulen < sizeof(*uh))
808 			goto short_packet;
809 
810 		if (ulen < skb->len) {
811 			if (pskb_trim_rcsum(skb, ulen))
812 				goto short_packet;
813 			saddr = &ipv6_hdr(skb)->saddr;
814 			daddr = &ipv6_hdr(skb)->daddr;
815 			uh = udp_hdr(skb);
816 		}
817 	}
818 
819 	if (udp6_csum_init(skb, uh, proto))
820 		goto discard;
821 
822 	/*
823 	 *	Multicast receive code
824 	 */
825 	if (ipv6_addr_is_multicast(daddr))
826 		return __udp6_lib_mcast_deliver(net, skb,
827 				saddr, daddr, udptable);
828 
829 	/* Unicast */
830 
831 	/*
832 	 * check socket cache ... must talk to Alan about his plans
833 	 * for sock caches... i'll skip this for now.
834 	 */
835 	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
836 	if (sk != NULL) {
837 		int ret = udpv6_queue_rcv_skb(sk, skb);
838 		sock_put(sk);
839 
840 		/* a return value > 0 means to resubmit the input, but
841 		 * it wants the return to be -protocol, or 0
842 		 */
843 		if (ret > 0)
844 			return -ret;
845 
846 		return 0;
847 	}
848 
849 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
850 		goto discard;
851 
852 	if (udp_lib_checksum_complete(skb))
853 		goto discard;
854 
855 	UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
856 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
857 
858 	kfree_skb(skb);
859 	return 0;
860 
861 short_packet:
862 	LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
863 		       proto == IPPROTO_UDPLITE ? "-Lite" : "",
864 		       saddr,
865 		       ntohs(uh->source),
866 		       ulen,
867 		       skb->len,
868 		       daddr,
869 		       ntohs(uh->dest));
870 
871 discard:
872 	UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
873 	kfree_skb(skb);
874 	return 0;
875 }
876 
877 static __inline__ int udpv6_rcv(struct sk_buff *skb)
878 {
879 	return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
880 }
881 
882 /*
883  * Throw away all pending data and cancel the corking. Socket is locked.
884  */
885 static void udp_v6_flush_pending_frames(struct sock *sk)
886 {
887 	struct udp_sock *up = udp_sk(sk);
888 
889 	if (up->pending == AF_INET)
890 		udp_flush_pending_frames(sk);
891 	else if (up->pending) {
892 		up->len = 0;
893 		up->pending = 0;
894 		ip6_flush_pending_frames(sk);
895 	}
896 }
897 
898 /**
899  * 	udp6_hwcsum_outgoing  -  handle outgoing HW checksumming
900  * 	@sk: 	socket we are sending on
901  * 	@skb: 	sk_buff containing the filled-in UDP header
902  * 	        (checksum field must be zeroed out)
903  */
904 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
905 				 const struct in6_addr *saddr,
906 				 const struct in6_addr *daddr, int len)
907 {
908 	unsigned int offset;
909 	struct udphdr *uh = udp_hdr(skb);
910 	__wsum csum = 0;
911 
912 	if (skb_queue_len(&sk->sk_write_queue) == 1) {
913 		/* Only one fragment on the socket.  */
914 		skb->csum_start = skb_transport_header(skb) - skb->head;
915 		skb->csum_offset = offsetof(struct udphdr, check);
916 		uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
917 	} else {
918 		/*
919 		 * HW-checksum won't work as there are two or more
920 		 * fragments on the socket so that all csums of sk_buffs
921 		 * should be together
922 		 */
923 		offset = skb_transport_offset(skb);
924 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
925 
926 		skb->ip_summed = CHECKSUM_NONE;
927 
928 		skb_queue_walk(&sk->sk_write_queue, skb) {
929 			csum = csum_add(csum, skb->csum);
930 		}
931 
932 		uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
933 					    csum);
934 		if (uh->check == 0)
935 			uh->check = CSUM_MANGLED_0;
936 	}
937 }
938 
939 /*
940  *	Sending
941  */
942 
943 static int udp_v6_push_pending_frames(struct sock *sk)
944 {
945 	struct sk_buff *skb;
946 	struct udphdr *uh;
947 	struct udp_sock  *up = udp_sk(sk);
948 	struct inet_sock *inet = inet_sk(sk);
949 	struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
950 	int err = 0;
951 	int is_udplite = IS_UDPLITE(sk);
952 	__wsum csum = 0;
953 
954 	/* Grab the skbuff where UDP header space exists. */
955 	if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
956 		goto out;
957 
958 	/*
959 	 * Create a UDP header
960 	 */
961 	uh = udp_hdr(skb);
962 	uh->source = fl6->fl6_sport;
963 	uh->dest = fl6->fl6_dport;
964 	uh->len = htons(up->len);
965 	uh->check = 0;
966 
967 	if (is_udplite)
968 		csum = udplite_csum_outgoing(sk, skb);
969 	else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
970 		udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr,
971 				     up->len);
972 		goto send;
973 	} else
974 		csum = udp_csum_outgoing(sk, skb);
975 
976 	/* add protocol-dependent pseudo-header */
977 	uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
978 				    up->len, fl6->flowi6_proto, csum);
979 	if (uh->check == 0)
980 		uh->check = CSUM_MANGLED_0;
981 
982 send:
983 	err = ip6_push_pending_frames(sk);
984 	if (err) {
985 		if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
986 			UDP6_INC_STATS_USER(sock_net(sk),
987 					    UDP_MIB_SNDBUFERRORS, is_udplite);
988 			err = 0;
989 		}
990 	} else
991 		UDP6_INC_STATS_USER(sock_net(sk),
992 				    UDP_MIB_OUTDATAGRAMS, is_udplite);
993 out:
994 	up->len = 0;
995 	up->pending = 0;
996 	return err;
997 }
998 
999 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
1000 		  struct msghdr *msg, size_t len)
1001 {
1002 	struct ipv6_txoptions opt_space;
1003 	struct udp_sock *up = udp_sk(sk);
1004 	struct inet_sock *inet = inet_sk(sk);
1005 	struct ipv6_pinfo *np = inet6_sk(sk);
1006 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
1007 	struct in6_addr *daddr, *final_p, final;
1008 	struct ipv6_txoptions *opt = NULL;
1009 	struct ip6_flowlabel *flowlabel = NULL;
1010 	struct flowi6 fl6;
1011 	struct dst_entry *dst;
1012 	int addr_len = msg->msg_namelen;
1013 	int ulen = len;
1014 	int hlimit = -1;
1015 	int tclass = -1;
1016 	int dontfrag = -1;
1017 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1018 	int err;
1019 	int connected = 0;
1020 	int is_udplite = IS_UDPLITE(sk);
1021 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1022 
1023 	/* destination address check */
1024 	if (sin6) {
1025 		if (addr_len < offsetof(struct sockaddr, sa_data))
1026 			return -EINVAL;
1027 
1028 		switch (sin6->sin6_family) {
1029 		case AF_INET6:
1030 			if (addr_len < SIN6_LEN_RFC2133)
1031 				return -EINVAL;
1032 			daddr = &sin6->sin6_addr;
1033 			break;
1034 		case AF_INET:
1035 			goto do_udp_sendmsg;
1036 		case AF_UNSPEC:
1037 			msg->msg_name = sin6 = NULL;
1038 			msg->msg_namelen = addr_len = 0;
1039 			daddr = NULL;
1040 			break;
1041 		default:
1042 			return -EINVAL;
1043 		}
1044 	} else if (!up->pending) {
1045 		if (sk->sk_state != TCP_ESTABLISHED)
1046 			return -EDESTADDRREQ;
1047 		daddr = &np->daddr;
1048 	} else
1049 		daddr = NULL;
1050 
1051 	if (daddr) {
1052 		if (ipv6_addr_v4mapped(daddr)) {
1053 			struct sockaddr_in sin;
1054 			sin.sin_family = AF_INET;
1055 			sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1056 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
1057 			msg->msg_name = &sin;
1058 			msg->msg_namelen = sizeof(sin);
1059 do_udp_sendmsg:
1060 			if (__ipv6_only_sock(sk))
1061 				return -ENETUNREACH;
1062 			return udp_sendmsg(iocb, sk, msg, len);
1063 		}
1064 	}
1065 
1066 	if (up->pending == AF_INET)
1067 		return udp_sendmsg(iocb, sk, msg, len);
1068 
1069 	/* Rough check on arithmetic overflow,
1070 	   better check is made in ip6_append_data().
1071 	   */
1072 	if (len > INT_MAX - sizeof(struct udphdr))
1073 		return -EMSGSIZE;
1074 
1075 	if (up->pending) {
1076 		/*
1077 		 * There are pending frames.
1078 		 * The socket lock must be held while it's corked.
1079 		 */
1080 		lock_sock(sk);
1081 		if (likely(up->pending)) {
1082 			if (unlikely(up->pending != AF_INET6)) {
1083 				release_sock(sk);
1084 				return -EAFNOSUPPORT;
1085 			}
1086 			dst = NULL;
1087 			goto do_append_data;
1088 		}
1089 		release_sock(sk);
1090 	}
1091 	ulen += sizeof(struct udphdr);
1092 
1093 	memset(&fl6, 0, sizeof(fl6));
1094 
1095 	if (sin6) {
1096 		if (sin6->sin6_port == 0)
1097 			return -EINVAL;
1098 
1099 		fl6.fl6_dport = sin6->sin6_port;
1100 		daddr = &sin6->sin6_addr;
1101 
1102 		if (np->sndflow) {
1103 			fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1104 			if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1105 				flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1106 				if (flowlabel == NULL)
1107 					return -EINVAL;
1108 				daddr = &flowlabel->dst;
1109 			}
1110 		}
1111 
1112 		/*
1113 		 * Otherwise it will be difficult to maintain
1114 		 * sk->sk_dst_cache.
1115 		 */
1116 		if (sk->sk_state == TCP_ESTABLISHED &&
1117 		    ipv6_addr_equal(daddr, &np->daddr))
1118 			daddr = &np->daddr;
1119 
1120 		if (addr_len >= sizeof(struct sockaddr_in6) &&
1121 		    sin6->sin6_scope_id &&
1122 		    __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1123 			fl6.flowi6_oif = sin6->sin6_scope_id;
1124 	} else {
1125 		if (sk->sk_state != TCP_ESTABLISHED)
1126 			return -EDESTADDRREQ;
1127 
1128 		fl6.fl6_dport = inet->inet_dport;
1129 		daddr = &np->daddr;
1130 		fl6.flowlabel = np->flow_label;
1131 		connected = 1;
1132 	}
1133 
1134 	if (!fl6.flowi6_oif)
1135 		fl6.flowi6_oif = sk->sk_bound_dev_if;
1136 
1137 	if (!fl6.flowi6_oif)
1138 		fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1139 
1140 	fl6.flowi6_mark = sk->sk_mark;
1141 
1142 	if (msg->msg_controllen) {
1143 		opt = &opt_space;
1144 		memset(opt, 0, sizeof(struct ipv6_txoptions));
1145 		opt->tot_len = sizeof(*opt);
1146 
1147 		err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1148 					    &hlimit, &tclass, &dontfrag);
1149 		if (err < 0) {
1150 			fl6_sock_release(flowlabel);
1151 			return err;
1152 		}
1153 		if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1154 			flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1155 			if (flowlabel == NULL)
1156 				return -EINVAL;
1157 		}
1158 		if (!(opt->opt_nflen|opt->opt_flen))
1159 			opt = NULL;
1160 		connected = 0;
1161 	}
1162 	if (opt == NULL)
1163 		opt = np->opt;
1164 	if (flowlabel)
1165 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
1166 	opt = ipv6_fixup_options(&opt_space, opt);
1167 
1168 	fl6.flowi6_proto = sk->sk_protocol;
1169 	if (!ipv6_addr_any(daddr))
1170 		fl6.daddr = *daddr;
1171 	else
1172 		fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1173 	if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1174 		fl6.saddr = np->saddr;
1175 	fl6.fl6_sport = inet->inet_sport;
1176 
1177 	final_p = fl6_update_dst(&fl6, opt, &final);
1178 	if (final_p)
1179 		connected = 0;
1180 
1181 	if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1182 		fl6.flowi6_oif = np->mcast_oif;
1183 		connected = 0;
1184 	} else if (!fl6.flowi6_oif)
1185 		fl6.flowi6_oif = np->ucast_oif;
1186 
1187 	security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1188 
1189 	dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, true);
1190 	if (IS_ERR(dst)) {
1191 		err = PTR_ERR(dst);
1192 		dst = NULL;
1193 		goto out;
1194 	}
1195 
1196 	if (hlimit < 0) {
1197 		if (ipv6_addr_is_multicast(&fl6.daddr))
1198 			hlimit = np->mcast_hops;
1199 		else
1200 			hlimit = np->hop_limit;
1201 		if (hlimit < 0)
1202 			hlimit = ip6_dst_hoplimit(dst);
1203 	}
1204 
1205 	if (tclass < 0)
1206 		tclass = np->tclass;
1207 
1208 	if (dontfrag < 0)
1209 		dontfrag = np->dontfrag;
1210 
1211 	if (msg->msg_flags&MSG_CONFIRM)
1212 		goto do_confirm;
1213 back_from_confirm:
1214 
1215 	lock_sock(sk);
1216 	if (unlikely(up->pending)) {
1217 		/* The socket is already corked while preparing it. */
1218 		/* ... which is an evident application bug. --ANK */
1219 		release_sock(sk);
1220 
1221 		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1222 		err = -EINVAL;
1223 		goto out;
1224 	}
1225 
1226 	up->pending = AF_INET6;
1227 
1228 do_append_data:
1229 	up->len += ulen;
1230 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
1231 	err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1232 		sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1233 		(struct rt6_info*)dst,
1234 		corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1235 	if (err)
1236 		udp_v6_flush_pending_frames(sk);
1237 	else if (!corkreq)
1238 		err = udp_v6_push_pending_frames(sk);
1239 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1240 		up->pending = 0;
1241 
1242 	if (dst) {
1243 		if (connected) {
1244 			ip6_dst_store(sk, dst,
1245 				      ipv6_addr_equal(&fl6.daddr, &np->daddr) ?
1246 				      &np->daddr : NULL,
1247 #ifdef CONFIG_IPV6_SUBTREES
1248 				      ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1249 				      &np->saddr :
1250 #endif
1251 				      NULL);
1252 		} else {
1253 			dst_release(dst);
1254 		}
1255 		dst = NULL;
1256 	}
1257 
1258 	if (err > 0)
1259 		err = np->recverr ? net_xmit_errno(err) : 0;
1260 	release_sock(sk);
1261 out:
1262 	dst_release(dst);
1263 	fl6_sock_release(flowlabel);
1264 	if (!err)
1265 		return len;
1266 	/*
1267 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1268 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1269 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1270 	 * things).  We could add another new stat but at least for now that
1271 	 * seems like overkill.
1272 	 */
1273 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1274 		UDP6_INC_STATS_USER(sock_net(sk),
1275 				UDP_MIB_SNDBUFERRORS, is_udplite);
1276 	}
1277 	return err;
1278 
1279 do_confirm:
1280 	dst_confirm(dst);
1281 	if (!(msg->msg_flags&MSG_PROBE) || len)
1282 		goto back_from_confirm;
1283 	err = 0;
1284 	goto out;
1285 }
1286 
1287 void udpv6_destroy_sock(struct sock *sk)
1288 {
1289 	struct udp_sock *up = udp_sk(sk);
1290 	lock_sock(sk);
1291 	udp_v6_flush_pending_frames(sk);
1292 	release_sock(sk);
1293 
1294 	if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1295 		void (*encap_destroy)(struct sock *sk);
1296 		encap_destroy = ACCESS_ONCE(up->encap_destroy);
1297 		if (encap_destroy)
1298 			encap_destroy(sk);
1299 	}
1300 
1301 	inet6_destroy_sock(sk);
1302 }
1303 
1304 /*
1305  *	Socket option code for UDP
1306  */
1307 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1308 		     char __user *optval, unsigned int optlen)
1309 {
1310 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1311 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1312 					  udp_v6_push_pending_frames);
1313 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
1314 }
1315 
1316 #ifdef CONFIG_COMPAT
1317 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1318 			    char __user *optval, unsigned int optlen)
1319 {
1320 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1321 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1322 					  udp_v6_push_pending_frames);
1323 	return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1324 }
1325 #endif
1326 
1327 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1328 		     char __user *optval, int __user *optlen)
1329 {
1330 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1331 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1332 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
1333 }
1334 
1335 #ifdef CONFIG_COMPAT
1336 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1337 			    char __user *optval, int __user *optlen)
1338 {
1339 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1340 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1341 	return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1342 }
1343 #endif
1344 
1345 static const struct inet6_protocol udpv6_protocol = {
1346 	.handler	=	udpv6_rcv,
1347 	.err_handler	=	udpv6_err,
1348 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1349 };
1350 
1351 /* ------------------------------------------------------------------------ */
1352 #ifdef CONFIG_PROC_FS
1353 
1354 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
1355 {
1356 	struct inet_sock *inet = inet_sk(sp);
1357 	struct ipv6_pinfo *np = inet6_sk(sp);
1358 	const struct in6_addr *dest, *src;
1359 	__u16 destp, srcp;
1360 
1361 	dest  = &np->daddr;
1362 	src   = &np->rcv_saddr;
1363 	destp = ntohs(inet->inet_dport);
1364 	srcp  = ntohs(inet->inet_sport);
1365 	seq_printf(seq,
1366 		   "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1367 		   "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n",
1368 		   bucket,
1369 		   src->s6_addr32[0], src->s6_addr32[1],
1370 		   src->s6_addr32[2], src->s6_addr32[3], srcp,
1371 		   dest->s6_addr32[0], dest->s6_addr32[1],
1372 		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
1373 		   sp->sk_state,
1374 		   sk_wmem_alloc_get(sp),
1375 		   sk_rmem_alloc_get(sp),
1376 		   0, 0L, 0,
1377 		   from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1378 		   0,
1379 		   sock_i_ino(sp),
1380 		   atomic_read(&sp->sk_refcnt), sp,
1381 		   atomic_read(&sp->sk_drops));
1382 }
1383 
1384 int udp6_seq_show(struct seq_file *seq, void *v)
1385 {
1386 	if (v == SEQ_START_TOKEN)
1387 		seq_printf(seq,
1388 			   "  sl  "
1389 			   "local_address                         "
1390 			   "remote_address                        "
1391 			   "st tx_queue rx_queue tr tm->when retrnsmt"
1392 			   "   uid  timeout inode ref pointer drops\n");
1393 	else
1394 		udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1395 	return 0;
1396 }
1397 
1398 static const struct file_operations udp6_afinfo_seq_fops = {
1399 	.owner    = THIS_MODULE,
1400 	.open     = udp_seq_open,
1401 	.read     = seq_read,
1402 	.llseek   = seq_lseek,
1403 	.release  = seq_release_net
1404 };
1405 
1406 static struct udp_seq_afinfo udp6_seq_afinfo = {
1407 	.name		= "udp6",
1408 	.family		= AF_INET6,
1409 	.udp_table	= &udp_table,
1410 	.seq_fops	= &udp6_afinfo_seq_fops,
1411 	.seq_ops	= {
1412 		.show		= udp6_seq_show,
1413 	},
1414 };
1415 
1416 int __net_init udp6_proc_init(struct net *net)
1417 {
1418 	return udp_proc_register(net, &udp6_seq_afinfo);
1419 }
1420 
1421 void udp6_proc_exit(struct net *net) {
1422 	udp_proc_unregister(net, &udp6_seq_afinfo);
1423 }
1424 #endif /* CONFIG_PROC_FS */
1425 
1426 /* ------------------------------------------------------------------------ */
1427 
1428 struct proto udpv6_prot = {
1429 	.name		   = "UDPv6",
1430 	.owner		   = THIS_MODULE,
1431 	.close		   = udp_lib_close,
1432 	.connect	   = ip6_datagram_connect,
1433 	.disconnect	   = udp_disconnect,
1434 	.ioctl		   = udp_ioctl,
1435 	.destroy	   = udpv6_destroy_sock,
1436 	.setsockopt	   = udpv6_setsockopt,
1437 	.getsockopt	   = udpv6_getsockopt,
1438 	.sendmsg	   = udpv6_sendmsg,
1439 	.recvmsg	   = udpv6_recvmsg,
1440 	.backlog_rcv	   = __udpv6_queue_rcv_skb,
1441 	.hash		   = udp_lib_hash,
1442 	.unhash		   = udp_lib_unhash,
1443 	.rehash		   = udp_v6_rehash,
1444 	.get_port	   = udp_v6_get_port,
1445 	.memory_allocated  = &udp_memory_allocated,
1446 	.sysctl_mem	   = sysctl_udp_mem,
1447 	.sysctl_wmem	   = &sysctl_udp_wmem_min,
1448 	.sysctl_rmem	   = &sysctl_udp_rmem_min,
1449 	.obj_size	   = sizeof(struct udp6_sock),
1450 	.slab_flags	   = SLAB_DESTROY_BY_RCU,
1451 	.h.udp_table	   = &udp_table,
1452 #ifdef CONFIG_COMPAT
1453 	.compat_setsockopt = compat_udpv6_setsockopt,
1454 	.compat_getsockopt = compat_udpv6_getsockopt,
1455 #endif
1456 	.clear_sk	   = sk_prot_clear_portaddr_nulls,
1457 };
1458 
1459 static struct inet_protosw udpv6_protosw = {
1460 	.type =      SOCK_DGRAM,
1461 	.protocol =  IPPROTO_UDP,
1462 	.prot =      &udpv6_prot,
1463 	.ops =       &inet6_dgram_ops,
1464 	.no_check =  UDP_CSUM_DEFAULT,
1465 	.flags =     INET_PROTOSW_PERMANENT,
1466 };
1467 
1468 
1469 int __init udpv6_init(void)
1470 {
1471 	int ret;
1472 
1473 	ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1474 	if (ret)
1475 		goto out;
1476 
1477 	ret = inet6_register_protosw(&udpv6_protosw);
1478 	if (ret)
1479 		goto out_udpv6_protocol;
1480 out:
1481 	return ret;
1482 
1483 out_udpv6_protocol:
1484 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1485 	goto out;
1486 }
1487 
1488 void udpv6_exit(void)
1489 {
1490 	inet6_unregister_protosw(&udpv6_protosw);
1491 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1492 }
1493