xref: /openbmc/linux/net/ipv6/udp.c (revision 52fb57e7)
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 #include <net/busy_poll.h>
50 
51 #include <linux/proc_fs.h>
52 #include <linux/seq_file.h>
53 #include <trace/events/skb.h>
54 #include "udp_impl.h"
55 
56 static u32 udp6_ehashfn(const struct net *net,
57 			const struct in6_addr *laddr,
58 			const u16 lport,
59 			const struct in6_addr *faddr,
60 			const __be16 fport)
61 {
62 	static u32 udp6_ehash_secret __read_mostly;
63 	static u32 udp_ipv6_hash_secret __read_mostly;
64 
65 	u32 lhash, fhash;
66 
67 	net_get_random_once(&udp6_ehash_secret,
68 			    sizeof(udp6_ehash_secret));
69 	net_get_random_once(&udp_ipv6_hash_secret,
70 			    sizeof(udp_ipv6_hash_secret));
71 
72 	lhash = (__force u32)laddr->s6_addr32[3];
73 	fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
74 
75 	return __inet6_ehashfn(lhash, lport, fhash, fport,
76 			       udp_ipv6_hash_secret + net_hash_mix(net));
77 }
78 
79 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
80 {
81 	const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
82 	int sk2_ipv6only = inet_v6_ipv6only(sk2);
83 	int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
84 	int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
85 
86 	/* if both are mapped, treat as IPv4 */
87 	if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
88 		return (!sk2_ipv6only &&
89 			(!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr ||
90 			  sk->sk_rcv_saddr == sk2->sk_rcv_saddr));
91 
92 	if (addr_type2 == IPV6_ADDR_ANY &&
93 	    !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
94 		return 1;
95 
96 	if (addr_type == IPV6_ADDR_ANY &&
97 	    !(ipv6_only_sock(sk) && addr_type2 == IPV6_ADDR_MAPPED))
98 		return 1;
99 
100 	if (sk2_rcv_saddr6 &&
101 	    ipv6_addr_equal(&sk->sk_v6_rcv_saddr, sk2_rcv_saddr6))
102 		return 1;
103 
104 	return 0;
105 }
106 
107 static u32 udp6_portaddr_hash(const struct net *net,
108 			      const struct in6_addr *addr6,
109 			      unsigned int port)
110 {
111 	unsigned int hash, mix = net_hash_mix(net);
112 
113 	if (ipv6_addr_any(addr6))
114 		hash = jhash_1word(0, mix);
115 	else if (ipv6_addr_v4mapped(addr6))
116 		hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
117 	else
118 		hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
119 
120 	return hash ^ port;
121 }
122 
123 int udp_v6_get_port(struct sock *sk, unsigned short snum)
124 {
125 	unsigned int hash2_nulladdr =
126 		udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
127 	unsigned int hash2_partial =
128 		udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
129 
130 	/* precompute partial secondary hash */
131 	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
132 	return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
133 }
134 
135 static void udp_v6_rehash(struct sock *sk)
136 {
137 	u16 new_hash = udp6_portaddr_hash(sock_net(sk),
138 					  &sk->sk_v6_rcv_saddr,
139 					  inet_sk(sk)->inet_num);
140 
141 	udp_lib_rehash(sk, new_hash);
142 }
143 
144 static inline int compute_score(struct sock *sk, struct net *net,
145 				unsigned short hnum,
146 				const struct in6_addr *saddr, __be16 sport,
147 				const struct in6_addr *daddr, __be16 dport,
148 				int dif)
149 {
150 	int score;
151 	struct inet_sock *inet;
152 
153 	if (!net_eq(sock_net(sk), net) ||
154 	    udp_sk(sk)->udp_port_hash != hnum ||
155 	    sk->sk_family != PF_INET6)
156 		return -1;
157 
158 	score = 0;
159 	inet = inet_sk(sk);
160 
161 	if (inet->inet_dport) {
162 		if (inet->inet_dport != sport)
163 			return -1;
164 		score++;
165 	}
166 
167 	if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
168 		if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
169 			return -1;
170 		score++;
171 	}
172 
173 	if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
174 		if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
175 			return -1;
176 		score++;
177 	}
178 
179 	if (sk->sk_bound_dev_if) {
180 		if (sk->sk_bound_dev_if != dif)
181 			return -1;
182 		score++;
183 	}
184 
185 	return score;
186 }
187 
188 #define SCORE2_MAX (1 + 1 + 1)
189 static inline int compute_score2(struct sock *sk, struct net *net,
190 				 const struct in6_addr *saddr, __be16 sport,
191 				 const struct in6_addr *daddr,
192 				 unsigned short hnum, int dif)
193 {
194 	int score;
195 	struct inet_sock *inet;
196 
197 	if (!net_eq(sock_net(sk), net) ||
198 	    udp_sk(sk)->udp_port_hash != hnum ||
199 	    sk->sk_family != PF_INET6)
200 		return -1;
201 
202 	if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
203 		return -1;
204 
205 	score = 0;
206 	inet = inet_sk(sk);
207 
208 	if (inet->inet_dport) {
209 		if (inet->inet_dport != sport)
210 			return -1;
211 		score++;
212 	}
213 
214 	if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
215 		if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
216 			return -1;
217 		score++;
218 	}
219 
220 	if (sk->sk_bound_dev_if) {
221 		if (sk->sk_bound_dev_if != dif)
222 			return -1;
223 		score++;
224 	}
225 
226 	return score;
227 }
228 
229 /* called with read_rcu_lock() */
230 static struct sock *udp6_lib_lookup2(struct net *net,
231 		const struct in6_addr *saddr, __be16 sport,
232 		const struct in6_addr *daddr, unsigned int hnum, int dif,
233 		struct udp_hslot *hslot2, unsigned int slot2)
234 {
235 	struct sock *sk, *result;
236 	struct hlist_nulls_node *node;
237 	int score, badness, matches = 0, reuseport = 0;
238 	u32 hash = 0;
239 
240 begin:
241 	result = NULL;
242 	badness = -1;
243 	udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
244 		score = compute_score2(sk, net, saddr, sport,
245 				      daddr, hnum, dif);
246 		if (score > badness) {
247 			result = sk;
248 			badness = score;
249 			reuseport = sk->sk_reuseport;
250 			if (reuseport) {
251 				hash = udp6_ehashfn(net, daddr, hnum,
252 						    saddr, sport);
253 				matches = 1;
254 			} else if (score == SCORE2_MAX)
255 				goto exact_match;
256 		} else if (score == badness && reuseport) {
257 			matches++;
258 			if (reciprocal_scale(hash, matches) == 0)
259 				result = sk;
260 			hash = next_pseudo_random32(hash);
261 		}
262 	}
263 	/*
264 	 * if the nulls value we got at the end of this lookup is
265 	 * not the expected one, we must restart lookup.
266 	 * We probably met an item that was moved to another chain.
267 	 */
268 	if (get_nulls_value(node) != slot2)
269 		goto begin;
270 
271 	if (result) {
272 exact_match:
273 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
274 			result = NULL;
275 		else if (unlikely(compute_score2(result, net, saddr, sport,
276 				  daddr, hnum, dif) < badness)) {
277 			sock_put(result);
278 			goto begin;
279 		}
280 	}
281 	return result;
282 }
283 
284 struct sock *__udp6_lib_lookup(struct net *net,
285 				      const struct in6_addr *saddr, __be16 sport,
286 				      const struct in6_addr *daddr, __be16 dport,
287 				      int dif, struct udp_table *udptable)
288 {
289 	struct sock *sk, *result;
290 	struct hlist_nulls_node *node;
291 	unsigned short hnum = ntohs(dport);
292 	unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
293 	struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
294 	int score, badness, matches = 0, reuseport = 0;
295 	u32 hash = 0;
296 
297 	rcu_read_lock();
298 	if (hslot->count > 10) {
299 		hash2 = udp6_portaddr_hash(net, daddr, hnum);
300 		slot2 = hash2 & udptable->mask;
301 		hslot2 = &udptable->hash2[slot2];
302 		if (hslot->count < hslot2->count)
303 			goto begin;
304 
305 		result = udp6_lib_lookup2(net, saddr, sport,
306 					  daddr, hnum, dif,
307 					  hslot2, slot2);
308 		if (!result) {
309 			hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
310 			slot2 = hash2 & udptable->mask;
311 			hslot2 = &udptable->hash2[slot2];
312 			if (hslot->count < hslot2->count)
313 				goto begin;
314 
315 			result = udp6_lib_lookup2(net, saddr, sport,
316 						  &in6addr_any, hnum, dif,
317 						  hslot2, slot2);
318 		}
319 		rcu_read_unlock();
320 		return result;
321 	}
322 begin:
323 	result = NULL;
324 	badness = -1;
325 	sk_nulls_for_each_rcu(sk, node, &hslot->head) {
326 		score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
327 		if (score > badness) {
328 			result = sk;
329 			badness = score;
330 			reuseport = sk->sk_reuseport;
331 			if (reuseport) {
332 				hash = udp6_ehashfn(net, daddr, hnum,
333 						    saddr, sport);
334 				matches = 1;
335 			}
336 		} else if (score == badness && reuseport) {
337 			matches++;
338 			if (reciprocal_scale(hash, matches) == 0)
339 				result = sk;
340 			hash = next_pseudo_random32(hash);
341 		}
342 	}
343 	/*
344 	 * if the nulls value we got at the end of this lookup is
345 	 * not the expected one, we must restart lookup.
346 	 * We probably met an item that was moved to another chain.
347 	 */
348 	if (get_nulls_value(node) != slot)
349 		goto begin;
350 
351 	if (result) {
352 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
353 			result = NULL;
354 		else if (unlikely(compute_score(result, net, hnum, saddr, sport,
355 					daddr, dport, dif) < badness)) {
356 			sock_put(result);
357 			goto begin;
358 		}
359 	}
360 	rcu_read_unlock();
361 	return result;
362 }
363 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
364 
365 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
366 					  __be16 sport, __be16 dport,
367 					  struct udp_table *udptable)
368 {
369 	struct sock *sk;
370 	const struct ipv6hdr *iph = ipv6_hdr(skb);
371 
372 	sk = skb_steal_sock(skb);
373 	if (unlikely(sk))
374 		return sk;
375 	return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
376 				 &iph->daddr, dport, inet6_iif(skb),
377 				 udptable);
378 }
379 
380 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
381 			     const struct in6_addr *daddr, __be16 dport, int dif)
382 {
383 	return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
384 }
385 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
386 
387 /*
388  *	This should be easy, if there is something there we
389  *	return it, otherwise we block.
390  */
391 
392 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
393 		  int noblock, int flags, int *addr_len)
394 {
395 	struct ipv6_pinfo *np = inet6_sk(sk);
396 	struct inet_sock *inet = inet_sk(sk);
397 	struct sk_buff *skb;
398 	unsigned int ulen, copied;
399 	int peeked, off = 0;
400 	int err;
401 	int is_udplite = IS_UDPLITE(sk);
402 	int is_udp4;
403 	bool slow;
404 
405 	if (flags & MSG_ERRQUEUE)
406 		return ipv6_recv_error(sk, msg, len, addr_len);
407 
408 	if (np->rxpmtu && np->rxopt.bits.rxpmtu)
409 		return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
410 
411 try_again:
412 	skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
413 				  &peeked, &off, &err);
414 	if (!skb)
415 		goto out;
416 
417 	ulen = skb->len - sizeof(struct udphdr);
418 	copied = len;
419 	if (copied > ulen)
420 		copied = ulen;
421 	else if (copied < ulen)
422 		msg->msg_flags |= MSG_TRUNC;
423 
424 	is_udp4 = (skb->protocol == htons(ETH_P_IP));
425 
426 	/*
427 	 * If checksum is needed at all, try to do it while copying the
428 	 * data.  If the data is truncated, or if we only want a partial
429 	 * coverage checksum (UDP-Lite), do it before the copy.
430 	 */
431 
432 	if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
433 		if (udp_lib_checksum_complete(skb))
434 			goto csum_copy_err;
435 	}
436 
437 	if (skb_csum_unnecessary(skb))
438 		err = skb_copy_datagram_msg(skb, sizeof(struct udphdr),
439 					    msg, copied);
440 	else {
441 		err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr), msg);
442 		if (err == -EINVAL)
443 			goto csum_copy_err;
444 	}
445 	if (unlikely(err)) {
446 		trace_kfree_skb(skb, udpv6_recvmsg);
447 		if (!peeked) {
448 			atomic_inc(&sk->sk_drops);
449 			if (is_udp4)
450 				UDP_INC_STATS_USER(sock_net(sk),
451 						   UDP_MIB_INERRORS,
452 						   is_udplite);
453 			else
454 				UDP6_INC_STATS_USER(sock_net(sk),
455 						    UDP_MIB_INERRORS,
456 						    is_udplite);
457 		}
458 		goto out_free;
459 	}
460 	if (!peeked) {
461 		if (is_udp4)
462 			UDP_INC_STATS_USER(sock_net(sk),
463 					UDP_MIB_INDATAGRAMS, is_udplite);
464 		else
465 			UDP6_INC_STATS_USER(sock_net(sk),
466 					UDP_MIB_INDATAGRAMS, is_udplite);
467 	}
468 
469 	sock_recv_ts_and_drops(msg, sk, skb);
470 
471 	/* Copy the address. */
472 	if (msg->msg_name) {
473 		DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
474 		sin6->sin6_family = AF_INET6;
475 		sin6->sin6_port = udp_hdr(skb)->source;
476 		sin6->sin6_flowinfo = 0;
477 
478 		if (is_udp4) {
479 			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
480 					       &sin6->sin6_addr);
481 			sin6->sin6_scope_id = 0;
482 		} else {
483 			sin6->sin6_addr = ipv6_hdr(skb)->saddr;
484 			sin6->sin6_scope_id =
485 				ipv6_iface_scope_id(&sin6->sin6_addr,
486 						    inet6_iif(skb));
487 		}
488 		*addr_len = sizeof(*sin6);
489 	}
490 
491 	if (np->rxopt.all)
492 		ip6_datagram_recv_common_ctl(sk, msg, skb);
493 
494 	if (is_udp4) {
495 		if (inet->cmsg_flags)
496 			ip_cmsg_recv(msg, skb);
497 	} else {
498 		if (np->rxopt.all)
499 			ip6_datagram_recv_specific_ctl(sk, msg, skb);
500 	}
501 
502 	err = copied;
503 	if (flags & MSG_TRUNC)
504 		err = ulen;
505 
506 out_free:
507 	skb_free_datagram_locked(sk, skb);
508 out:
509 	return err;
510 
511 csum_copy_err:
512 	slow = lock_sock_fast(sk);
513 	if (!skb_kill_datagram(sk, skb, flags)) {
514 		if (is_udp4) {
515 			UDP_INC_STATS_USER(sock_net(sk),
516 					UDP_MIB_CSUMERRORS, is_udplite);
517 			UDP_INC_STATS_USER(sock_net(sk),
518 					UDP_MIB_INERRORS, is_udplite);
519 		} else {
520 			UDP6_INC_STATS_USER(sock_net(sk),
521 					UDP_MIB_CSUMERRORS, is_udplite);
522 			UDP6_INC_STATS_USER(sock_net(sk),
523 					UDP_MIB_INERRORS, is_udplite);
524 		}
525 	}
526 	unlock_sock_fast(sk, slow);
527 
528 	if (noblock)
529 		return -EAGAIN;
530 
531 	/* starting over for a new packet */
532 	msg->msg_flags &= ~MSG_TRUNC;
533 	goto try_again;
534 }
535 
536 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
537 		    u8 type, u8 code, int offset, __be32 info,
538 		    struct udp_table *udptable)
539 {
540 	struct ipv6_pinfo *np;
541 	const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
542 	const struct in6_addr *saddr = &hdr->saddr;
543 	const struct in6_addr *daddr = &hdr->daddr;
544 	struct udphdr *uh = (struct udphdr *)(skb->data+offset);
545 	struct sock *sk;
546 	int err;
547 	struct net *net = dev_net(skb->dev);
548 
549 	sk = __udp6_lib_lookup(net, daddr, uh->dest,
550 			       saddr, uh->source, inet6_iif(skb), udptable);
551 	if (!sk) {
552 		ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
553 				   ICMP6_MIB_INERRORS);
554 		return;
555 	}
556 
557 	if (type == ICMPV6_PKT_TOOBIG) {
558 		if (!ip6_sk_accept_pmtu(sk))
559 			goto out;
560 		ip6_sk_update_pmtu(skb, sk, info);
561 	}
562 	if (type == NDISC_REDIRECT) {
563 		ip6_sk_redirect(skb, sk);
564 		goto out;
565 	}
566 
567 	np = inet6_sk(sk);
568 
569 	if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
570 		goto out;
571 
572 	if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
573 		goto out;
574 
575 	if (np->recverr)
576 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
577 
578 	sk->sk_err = err;
579 	sk->sk_error_report(sk);
580 out:
581 	sock_put(sk);
582 }
583 
584 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
585 {
586 	int rc;
587 
588 	if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
589 		sock_rps_save_rxhash(sk, skb);
590 		sk_mark_napi_id(sk, skb);
591 		sk_incoming_cpu_update(sk);
592 	}
593 
594 	rc = sock_queue_rcv_skb(sk, skb);
595 	if (rc < 0) {
596 		int is_udplite = IS_UDPLITE(sk);
597 
598 		/* Note that an ENOMEM error is charged twice */
599 		if (rc == -ENOMEM)
600 			UDP6_INC_STATS_BH(sock_net(sk),
601 					UDP_MIB_RCVBUFERRORS, is_udplite);
602 		UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
603 		kfree_skb(skb);
604 		return -1;
605 	}
606 	return 0;
607 }
608 
609 static __inline__ void udpv6_err(struct sk_buff *skb,
610 				 struct inet6_skb_parm *opt, u8 type,
611 				 u8 code, int offset, __be32 info)
612 {
613 	__udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
614 }
615 
616 static struct static_key udpv6_encap_needed __read_mostly;
617 void udpv6_encap_enable(void)
618 {
619 	if (!static_key_enabled(&udpv6_encap_needed))
620 		static_key_slow_inc(&udpv6_encap_needed);
621 }
622 EXPORT_SYMBOL(udpv6_encap_enable);
623 
624 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
625 {
626 	struct udp_sock *up = udp_sk(sk);
627 	int rc;
628 	int is_udplite = IS_UDPLITE(sk);
629 
630 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
631 		goto drop;
632 
633 	if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
634 		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
635 
636 		/*
637 		 * This is an encapsulation socket so pass the skb to
638 		 * the socket's udp_encap_rcv() hook. Otherwise, just
639 		 * fall through and pass this up the UDP socket.
640 		 * up->encap_rcv() returns the following value:
641 		 * =0 if skb was successfully passed to the encap
642 		 *    handler or was discarded by it.
643 		 * >0 if skb should be passed on to UDP.
644 		 * <0 if skb should be resubmitted as proto -N
645 		 */
646 
647 		/* if we're overly short, let UDP handle it */
648 		encap_rcv = ACCESS_ONCE(up->encap_rcv);
649 		if (skb->len > sizeof(struct udphdr) && encap_rcv) {
650 			int ret;
651 
652 			/* Verify checksum before giving to encap */
653 			if (udp_lib_checksum_complete(skb))
654 				goto csum_error;
655 
656 			ret = encap_rcv(sk, skb);
657 			if (ret <= 0) {
658 				UDP_INC_STATS_BH(sock_net(sk),
659 						 UDP_MIB_INDATAGRAMS,
660 						 is_udplite);
661 				return -ret;
662 			}
663 		}
664 
665 		/* FALLTHROUGH -- it's a UDP Packet */
666 	}
667 
668 	/*
669 	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
670 	 */
671 	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
672 
673 		if (up->pcrlen == 0) {          /* full coverage was set  */
674 			net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
675 					    UDP_SKB_CB(skb)->cscov, skb->len);
676 			goto drop;
677 		}
678 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
679 			net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
680 					    UDP_SKB_CB(skb)->cscov, up->pcrlen);
681 			goto drop;
682 		}
683 	}
684 
685 	if (rcu_access_pointer(sk->sk_filter)) {
686 		if (udp_lib_checksum_complete(skb))
687 			goto csum_error;
688 	}
689 
690 	if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
691 		UDP6_INC_STATS_BH(sock_net(sk),
692 				  UDP_MIB_RCVBUFERRORS, is_udplite);
693 		goto drop;
694 	}
695 
696 	skb_dst_drop(skb);
697 
698 	bh_lock_sock(sk);
699 	rc = 0;
700 	if (!sock_owned_by_user(sk))
701 		rc = __udpv6_queue_rcv_skb(sk, skb);
702 	else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
703 		bh_unlock_sock(sk);
704 		goto drop;
705 	}
706 	bh_unlock_sock(sk);
707 
708 	return rc;
709 
710 csum_error:
711 	UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
712 drop:
713 	UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
714 	atomic_inc(&sk->sk_drops);
715 	kfree_skb(skb);
716 	return -1;
717 }
718 
719 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
720 				   __be16 loc_port, const struct in6_addr *loc_addr,
721 				   __be16 rmt_port, const struct in6_addr *rmt_addr,
722 				   int dif, unsigned short hnum)
723 {
724 	struct inet_sock *inet = inet_sk(sk);
725 
726 	if (!net_eq(sock_net(sk), net))
727 		return false;
728 
729 	if (udp_sk(sk)->udp_port_hash != hnum ||
730 	    sk->sk_family != PF_INET6 ||
731 	    (inet->inet_dport && inet->inet_dport != rmt_port) ||
732 	    (!ipv6_addr_any(&sk->sk_v6_daddr) &&
733 		    !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
734 	    (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
735 	    (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
736 		    !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
737 		return false;
738 	if (!inet6_mc_check(sk, loc_addr, rmt_addr))
739 		return false;
740 	return true;
741 }
742 
743 static void flush_stack(struct sock **stack, unsigned int count,
744 			struct sk_buff *skb, unsigned int final)
745 {
746 	struct sk_buff *skb1 = NULL;
747 	struct sock *sk;
748 	unsigned int i;
749 
750 	for (i = 0; i < count; i++) {
751 		sk = stack[i];
752 		if (likely(!skb1))
753 			skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
754 		if (!skb1) {
755 			atomic_inc(&sk->sk_drops);
756 			UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
757 					  IS_UDPLITE(sk));
758 			UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
759 					  IS_UDPLITE(sk));
760 		}
761 
762 		if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
763 			skb1 = NULL;
764 		sock_put(sk);
765 	}
766 	if (unlikely(skb1))
767 		kfree_skb(skb1);
768 }
769 
770 static void udp6_csum_zero_error(struct sk_buff *skb)
771 {
772 	/* RFC 2460 section 8.1 says that we SHOULD log
773 	 * this error. Well, it is reasonable.
774 	 */
775 	net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
776 			    &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
777 			    &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
778 }
779 
780 /*
781  * Note: called only from the BH handler context,
782  * so we don't need to lock the hashes.
783  */
784 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
785 		const struct in6_addr *saddr, const struct in6_addr *daddr,
786 		struct udp_table *udptable, int proto)
787 {
788 	struct sock *sk, *stack[256 / sizeof(struct sock *)];
789 	const struct udphdr *uh = udp_hdr(skb);
790 	struct hlist_nulls_node *node;
791 	unsigned short hnum = ntohs(uh->dest);
792 	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
793 	int dif = inet6_iif(skb);
794 	unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
795 	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
796 	bool inner_flushed = false;
797 
798 	if (use_hash2) {
799 		hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
800 			    udp_table.mask;
801 		hash2 = udp6_portaddr_hash(net, daddr, hnum) & udp_table.mask;
802 start_lookup:
803 		hslot = &udp_table.hash2[hash2];
804 		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
805 	}
806 
807 	spin_lock(&hslot->lock);
808 	sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
809 		if (__udp_v6_is_mcast_sock(net, sk,
810 					   uh->dest, daddr,
811 					   uh->source, saddr,
812 					   dif, hnum) &&
813 		    /* If zero checksum and no_check is not on for
814 		     * the socket then skip it.
815 		     */
816 		    (uh->check || udp_sk(sk)->no_check6_rx)) {
817 			if (unlikely(count == ARRAY_SIZE(stack))) {
818 				flush_stack(stack, count, skb, ~0);
819 				inner_flushed = true;
820 				count = 0;
821 			}
822 			stack[count++] = sk;
823 			sock_hold(sk);
824 		}
825 	}
826 
827 	spin_unlock(&hslot->lock);
828 
829 	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
830 	if (use_hash2 && hash2 != hash2_any) {
831 		hash2 = hash2_any;
832 		goto start_lookup;
833 	}
834 
835 	if (count) {
836 		flush_stack(stack, count, skb, count - 1);
837 	} else {
838 		if (!inner_flushed)
839 			UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
840 					 proto == IPPROTO_UDPLITE);
841 		consume_skb(skb);
842 	}
843 	return 0;
844 }
845 
846 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
847 		   int proto)
848 {
849 	struct net *net = dev_net(skb->dev);
850 	struct sock *sk;
851 	struct udphdr *uh;
852 	const struct in6_addr *saddr, *daddr;
853 	u32 ulen = 0;
854 
855 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
856 		goto discard;
857 
858 	saddr = &ipv6_hdr(skb)->saddr;
859 	daddr = &ipv6_hdr(skb)->daddr;
860 	uh = udp_hdr(skb);
861 
862 	ulen = ntohs(uh->len);
863 	if (ulen > skb->len)
864 		goto short_packet;
865 
866 	if (proto == IPPROTO_UDP) {
867 		/* UDP validates ulen. */
868 
869 		/* Check for jumbo payload */
870 		if (ulen == 0)
871 			ulen = skb->len;
872 
873 		if (ulen < sizeof(*uh))
874 			goto short_packet;
875 
876 		if (ulen < skb->len) {
877 			if (pskb_trim_rcsum(skb, ulen))
878 				goto short_packet;
879 			saddr = &ipv6_hdr(skb)->saddr;
880 			daddr = &ipv6_hdr(skb)->daddr;
881 			uh = udp_hdr(skb);
882 		}
883 	}
884 
885 	if (udp6_csum_init(skb, uh, proto))
886 		goto csum_error;
887 
888 	/*
889 	 *	Multicast receive code
890 	 */
891 	if (ipv6_addr_is_multicast(daddr))
892 		return __udp6_lib_mcast_deliver(net, skb,
893 				saddr, daddr, udptable, proto);
894 
895 	/* Unicast */
896 
897 	/*
898 	 * check socket cache ... must talk to Alan about his plans
899 	 * for sock caches... i'll skip this for now.
900 	 */
901 	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
902 	if (sk) {
903 		int ret;
904 
905 		if (!uh->check && !udp_sk(sk)->no_check6_rx) {
906 			sock_put(sk);
907 			udp6_csum_zero_error(skb);
908 			goto csum_error;
909 		}
910 
911 		if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
912 			skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
913 						 ip6_compute_pseudo);
914 
915 		ret = udpv6_queue_rcv_skb(sk, skb);
916 		sock_put(sk);
917 
918 		/* a return value > 0 means to resubmit the input, but
919 		 * it wants the return to be -protocol, or 0
920 		 */
921 		if (ret > 0)
922 			return -ret;
923 
924 		return 0;
925 	}
926 
927 	if (!uh->check) {
928 		udp6_csum_zero_error(skb);
929 		goto csum_error;
930 	}
931 
932 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
933 		goto discard;
934 
935 	if (udp_lib_checksum_complete(skb))
936 		goto csum_error;
937 
938 	UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
939 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
940 
941 	kfree_skb(skb);
942 	return 0;
943 
944 short_packet:
945 	net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
946 			    proto == IPPROTO_UDPLITE ? "-Lite" : "",
947 			    saddr, ntohs(uh->source),
948 			    ulen, skb->len,
949 			    daddr, ntohs(uh->dest));
950 	goto discard;
951 csum_error:
952 	UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
953 discard:
954 	UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
955 	kfree_skb(skb);
956 	return 0;
957 }
958 
959 static __inline__ int udpv6_rcv(struct sk_buff *skb)
960 {
961 	return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
962 }
963 
964 /*
965  * Throw away all pending data and cancel the corking. Socket is locked.
966  */
967 static void udp_v6_flush_pending_frames(struct sock *sk)
968 {
969 	struct udp_sock *up = udp_sk(sk);
970 
971 	if (up->pending == AF_INET)
972 		udp_flush_pending_frames(sk);
973 	else if (up->pending) {
974 		up->len = 0;
975 		up->pending = 0;
976 		ip6_flush_pending_frames(sk);
977 	}
978 }
979 
980 /**
981  *	udp6_hwcsum_outgoing  -  handle outgoing HW checksumming
982  *	@sk:	socket we are sending on
983  *	@skb:	sk_buff containing the filled-in UDP header
984  *		(checksum field must be zeroed out)
985  */
986 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
987 				 const struct in6_addr *saddr,
988 				 const struct in6_addr *daddr, int len)
989 {
990 	unsigned int offset;
991 	struct udphdr *uh = udp_hdr(skb);
992 	struct sk_buff *frags = skb_shinfo(skb)->frag_list;
993 	__wsum csum = 0;
994 
995 	if (!frags) {
996 		/* Only one fragment on the socket.  */
997 		skb->csum_start = skb_transport_header(skb) - skb->head;
998 		skb->csum_offset = offsetof(struct udphdr, check);
999 		uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1000 	} else {
1001 		/*
1002 		 * HW-checksum won't work as there are two or more
1003 		 * fragments on the socket so that all csums of sk_buffs
1004 		 * should be together
1005 		 */
1006 		offset = skb_transport_offset(skb);
1007 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1008 
1009 		skb->ip_summed = CHECKSUM_NONE;
1010 
1011 		do {
1012 			csum = csum_add(csum, frags->csum);
1013 		} while ((frags = frags->next));
1014 
1015 		uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1016 					    csum);
1017 		if (uh->check == 0)
1018 			uh->check = CSUM_MANGLED_0;
1019 	}
1020 }
1021 
1022 /*
1023  *	Sending
1024  */
1025 
1026 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
1027 {
1028 	struct sock *sk = skb->sk;
1029 	struct udphdr *uh;
1030 	int err = 0;
1031 	int is_udplite = IS_UDPLITE(sk);
1032 	__wsum csum = 0;
1033 	int offset = skb_transport_offset(skb);
1034 	int len = skb->len - offset;
1035 
1036 	/*
1037 	 * Create a UDP header
1038 	 */
1039 	uh = udp_hdr(skb);
1040 	uh->source = fl6->fl6_sport;
1041 	uh->dest = fl6->fl6_dport;
1042 	uh->len = htons(len);
1043 	uh->check = 0;
1044 
1045 	if (is_udplite)
1046 		csum = udplite_csum(skb);
1047 	else if (udp_sk(sk)->no_check6_tx) {   /* UDP csum disabled */
1048 		skb->ip_summed = CHECKSUM_NONE;
1049 		goto send;
1050 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1051 		udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1052 		goto send;
1053 	} else
1054 		csum = udp_csum(skb);
1055 
1056 	/* add protocol-dependent pseudo-header */
1057 	uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1058 				    len, fl6->flowi6_proto, csum);
1059 	if (uh->check == 0)
1060 		uh->check = CSUM_MANGLED_0;
1061 
1062 send:
1063 	err = ip6_send_skb(skb);
1064 	if (err) {
1065 		if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1066 			UDP6_INC_STATS_USER(sock_net(sk),
1067 					    UDP_MIB_SNDBUFERRORS, is_udplite);
1068 			err = 0;
1069 		}
1070 	} else
1071 		UDP6_INC_STATS_USER(sock_net(sk),
1072 				    UDP_MIB_OUTDATAGRAMS, is_udplite);
1073 	return err;
1074 }
1075 
1076 static int udp_v6_push_pending_frames(struct sock *sk)
1077 {
1078 	struct sk_buff *skb;
1079 	struct udp_sock  *up = udp_sk(sk);
1080 	struct flowi6 fl6;
1081 	int err = 0;
1082 
1083 	if (up->pending == AF_INET)
1084 		return udp_push_pending_frames(sk);
1085 
1086 	/* ip6_finish_skb will release the cork, so make a copy of
1087 	 * fl6 here.
1088 	 */
1089 	fl6 = inet_sk(sk)->cork.fl.u.ip6;
1090 
1091 	skb = ip6_finish_skb(sk);
1092 	if (!skb)
1093 		goto out;
1094 
1095 	err = udp_v6_send_skb(skb, &fl6);
1096 
1097 out:
1098 	up->len = 0;
1099 	up->pending = 0;
1100 	return err;
1101 }
1102 
1103 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1104 {
1105 	struct ipv6_txoptions opt_space;
1106 	struct udp_sock *up = udp_sk(sk);
1107 	struct inet_sock *inet = inet_sk(sk);
1108 	struct ipv6_pinfo *np = inet6_sk(sk);
1109 	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1110 	struct in6_addr *daddr, *final_p, final;
1111 	struct ipv6_txoptions *opt = NULL;
1112 	struct ip6_flowlabel *flowlabel = NULL;
1113 	struct flowi6 fl6;
1114 	struct dst_entry *dst;
1115 	int addr_len = msg->msg_namelen;
1116 	int ulen = len;
1117 	int hlimit = -1;
1118 	int tclass = -1;
1119 	int dontfrag = -1;
1120 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1121 	int err;
1122 	int connected = 0;
1123 	int is_udplite = IS_UDPLITE(sk);
1124 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1125 
1126 	/* destination address check */
1127 	if (sin6) {
1128 		if (addr_len < offsetof(struct sockaddr, sa_data))
1129 			return -EINVAL;
1130 
1131 		switch (sin6->sin6_family) {
1132 		case AF_INET6:
1133 			if (addr_len < SIN6_LEN_RFC2133)
1134 				return -EINVAL;
1135 			daddr = &sin6->sin6_addr;
1136 			break;
1137 		case AF_INET:
1138 			goto do_udp_sendmsg;
1139 		case AF_UNSPEC:
1140 			msg->msg_name = sin6 = NULL;
1141 			msg->msg_namelen = addr_len = 0;
1142 			daddr = NULL;
1143 			break;
1144 		default:
1145 			return -EINVAL;
1146 		}
1147 	} else if (!up->pending) {
1148 		if (sk->sk_state != TCP_ESTABLISHED)
1149 			return -EDESTADDRREQ;
1150 		daddr = &sk->sk_v6_daddr;
1151 	} else
1152 		daddr = NULL;
1153 
1154 	if (daddr) {
1155 		if (ipv6_addr_v4mapped(daddr)) {
1156 			struct sockaddr_in sin;
1157 			sin.sin_family = AF_INET;
1158 			sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1159 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
1160 			msg->msg_name = &sin;
1161 			msg->msg_namelen = sizeof(sin);
1162 do_udp_sendmsg:
1163 			if (__ipv6_only_sock(sk))
1164 				return -ENETUNREACH;
1165 			return udp_sendmsg(sk, msg, len);
1166 		}
1167 	}
1168 
1169 	if (up->pending == AF_INET)
1170 		return udp_sendmsg(sk, msg, len);
1171 
1172 	/* Rough check on arithmetic overflow,
1173 	   better check is made in ip6_append_data().
1174 	   */
1175 	if (len > INT_MAX - sizeof(struct udphdr))
1176 		return -EMSGSIZE;
1177 
1178 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
1179 	if (up->pending) {
1180 		/*
1181 		 * There are pending frames.
1182 		 * The socket lock must be held while it's corked.
1183 		 */
1184 		lock_sock(sk);
1185 		if (likely(up->pending)) {
1186 			if (unlikely(up->pending != AF_INET6)) {
1187 				release_sock(sk);
1188 				return -EAFNOSUPPORT;
1189 			}
1190 			dst = NULL;
1191 			goto do_append_data;
1192 		}
1193 		release_sock(sk);
1194 	}
1195 	ulen += sizeof(struct udphdr);
1196 
1197 	memset(&fl6, 0, sizeof(fl6));
1198 
1199 	if (sin6) {
1200 		if (sin6->sin6_port == 0)
1201 			return -EINVAL;
1202 
1203 		fl6.fl6_dport = sin6->sin6_port;
1204 		daddr = &sin6->sin6_addr;
1205 
1206 		if (np->sndflow) {
1207 			fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1208 			if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1209 				flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1210 				if (!flowlabel)
1211 					return -EINVAL;
1212 			}
1213 		}
1214 
1215 		/*
1216 		 * Otherwise it will be difficult to maintain
1217 		 * sk->sk_dst_cache.
1218 		 */
1219 		if (sk->sk_state == TCP_ESTABLISHED &&
1220 		    ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1221 			daddr = &sk->sk_v6_daddr;
1222 
1223 		if (addr_len >= sizeof(struct sockaddr_in6) &&
1224 		    sin6->sin6_scope_id &&
1225 		    __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1226 			fl6.flowi6_oif = sin6->sin6_scope_id;
1227 	} else {
1228 		if (sk->sk_state != TCP_ESTABLISHED)
1229 			return -EDESTADDRREQ;
1230 
1231 		fl6.fl6_dport = inet->inet_dport;
1232 		daddr = &sk->sk_v6_daddr;
1233 		fl6.flowlabel = np->flow_label;
1234 		connected = 1;
1235 	}
1236 
1237 	if (!fl6.flowi6_oif)
1238 		fl6.flowi6_oif = sk->sk_bound_dev_if;
1239 
1240 	if (!fl6.flowi6_oif)
1241 		fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1242 
1243 	fl6.flowi6_mark = sk->sk_mark;
1244 
1245 	if (msg->msg_controllen) {
1246 		opt = &opt_space;
1247 		memset(opt, 0, sizeof(struct ipv6_txoptions));
1248 		opt->tot_len = sizeof(*opt);
1249 
1250 		err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1251 					    &hlimit, &tclass, &dontfrag);
1252 		if (err < 0) {
1253 			fl6_sock_release(flowlabel);
1254 			return err;
1255 		}
1256 		if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1257 			flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1258 			if (!flowlabel)
1259 				return -EINVAL;
1260 		}
1261 		if (!(opt->opt_nflen|opt->opt_flen))
1262 			opt = NULL;
1263 		connected = 0;
1264 	}
1265 	if (!opt)
1266 		opt = np->opt;
1267 	if (flowlabel)
1268 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
1269 	opt = ipv6_fixup_options(&opt_space, opt);
1270 
1271 	fl6.flowi6_proto = sk->sk_protocol;
1272 	if (!ipv6_addr_any(daddr))
1273 		fl6.daddr = *daddr;
1274 	else
1275 		fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1276 	if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1277 		fl6.saddr = np->saddr;
1278 	fl6.fl6_sport = inet->inet_sport;
1279 
1280 	final_p = fl6_update_dst(&fl6, opt, &final);
1281 	if (final_p)
1282 		connected = 0;
1283 
1284 	if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1285 		fl6.flowi6_oif = np->mcast_oif;
1286 		connected = 0;
1287 	} else if (!fl6.flowi6_oif)
1288 		fl6.flowi6_oif = np->ucast_oif;
1289 
1290 	security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1291 
1292 	dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1293 	if (IS_ERR(dst)) {
1294 		err = PTR_ERR(dst);
1295 		dst = NULL;
1296 		goto out;
1297 	}
1298 
1299 	if (hlimit < 0)
1300 		hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1301 
1302 	if (tclass < 0)
1303 		tclass = np->tclass;
1304 
1305 	if (msg->msg_flags&MSG_CONFIRM)
1306 		goto do_confirm;
1307 back_from_confirm:
1308 
1309 	/* Lockless fast path for the non-corking case */
1310 	if (!corkreq) {
1311 		struct sk_buff *skb;
1312 
1313 		skb = ip6_make_skb(sk, getfrag, msg, ulen,
1314 				   sizeof(struct udphdr), hlimit, tclass, opt,
1315 				   &fl6, (struct rt6_info *)dst,
1316 				   msg->msg_flags, dontfrag);
1317 		err = PTR_ERR(skb);
1318 		if (!IS_ERR_OR_NULL(skb))
1319 			err = udp_v6_send_skb(skb, &fl6);
1320 		goto release_dst;
1321 	}
1322 
1323 	lock_sock(sk);
1324 	if (unlikely(up->pending)) {
1325 		/* The socket is already corked while preparing it. */
1326 		/* ... which is an evident application bug. --ANK */
1327 		release_sock(sk);
1328 
1329 		net_dbg_ratelimited("udp cork app bug 2\n");
1330 		err = -EINVAL;
1331 		goto out;
1332 	}
1333 
1334 	up->pending = AF_INET6;
1335 
1336 do_append_data:
1337 	if (dontfrag < 0)
1338 		dontfrag = np->dontfrag;
1339 	up->len += ulen;
1340 	err = ip6_append_data(sk, getfrag, msg, ulen,
1341 		sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1342 		(struct rt6_info *)dst,
1343 		corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1344 	if (err)
1345 		udp_v6_flush_pending_frames(sk);
1346 	else if (!corkreq)
1347 		err = udp_v6_push_pending_frames(sk);
1348 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1349 		up->pending = 0;
1350 
1351 	if (err > 0)
1352 		err = np->recverr ? net_xmit_errno(err) : 0;
1353 	release_sock(sk);
1354 
1355 release_dst:
1356 	if (dst) {
1357 		if (connected) {
1358 			ip6_dst_store(sk, dst,
1359 				      ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1360 				      &sk->sk_v6_daddr : NULL,
1361 #ifdef CONFIG_IPV6_SUBTREES
1362 				      ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1363 				      &np->saddr :
1364 #endif
1365 				      NULL);
1366 		} else {
1367 			dst_release(dst);
1368 		}
1369 		dst = NULL;
1370 	}
1371 
1372 out:
1373 	dst_release(dst);
1374 	fl6_sock_release(flowlabel);
1375 	if (!err)
1376 		return len;
1377 	/*
1378 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1379 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1380 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1381 	 * things).  We could add another new stat but at least for now that
1382 	 * seems like overkill.
1383 	 */
1384 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1385 		UDP6_INC_STATS_USER(sock_net(sk),
1386 				UDP_MIB_SNDBUFERRORS, is_udplite);
1387 	}
1388 	return err;
1389 
1390 do_confirm:
1391 	dst_confirm(dst);
1392 	if (!(msg->msg_flags&MSG_PROBE) || len)
1393 		goto back_from_confirm;
1394 	err = 0;
1395 	goto out;
1396 }
1397 
1398 void udpv6_destroy_sock(struct sock *sk)
1399 {
1400 	struct udp_sock *up = udp_sk(sk);
1401 	lock_sock(sk);
1402 	udp_v6_flush_pending_frames(sk);
1403 	release_sock(sk);
1404 
1405 	if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1406 		void (*encap_destroy)(struct sock *sk);
1407 		encap_destroy = ACCESS_ONCE(up->encap_destroy);
1408 		if (encap_destroy)
1409 			encap_destroy(sk);
1410 	}
1411 
1412 	inet6_destroy_sock(sk);
1413 }
1414 
1415 /*
1416  *	Socket option code for UDP
1417  */
1418 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1419 		     char __user *optval, unsigned int optlen)
1420 {
1421 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1422 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1423 					  udp_v6_push_pending_frames);
1424 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
1425 }
1426 
1427 #ifdef CONFIG_COMPAT
1428 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1429 			    char __user *optval, unsigned int optlen)
1430 {
1431 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1432 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1433 					  udp_v6_push_pending_frames);
1434 	return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1435 }
1436 #endif
1437 
1438 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1439 		     char __user *optval, int __user *optlen)
1440 {
1441 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1442 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1443 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
1444 }
1445 
1446 #ifdef CONFIG_COMPAT
1447 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1448 			    char __user *optval, int __user *optlen)
1449 {
1450 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1451 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1452 	return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1453 }
1454 #endif
1455 
1456 static const struct inet6_protocol udpv6_protocol = {
1457 	.handler	=	udpv6_rcv,
1458 	.err_handler	=	udpv6_err,
1459 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1460 };
1461 
1462 /* ------------------------------------------------------------------------ */
1463 #ifdef CONFIG_PROC_FS
1464 int udp6_seq_show(struct seq_file *seq, void *v)
1465 {
1466 	if (v == SEQ_START_TOKEN) {
1467 		seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1468 	} else {
1469 		int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1470 		struct inet_sock *inet = inet_sk(v);
1471 		__u16 srcp = ntohs(inet->inet_sport);
1472 		__u16 destp = ntohs(inet->inet_dport);
1473 		ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1474 	}
1475 	return 0;
1476 }
1477 
1478 static const struct file_operations udp6_afinfo_seq_fops = {
1479 	.owner    = THIS_MODULE,
1480 	.open     = udp_seq_open,
1481 	.read     = seq_read,
1482 	.llseek   = seq_lseek,
1483 	.release  = seq_release_net
1484 };
1485 
1486 static struct udp_seq_afinfo udp6_seq_afinfo = {
1487 	.name		= "udp6",
1488 	.family		= AF_INET6,
1489 	.udp_table	= &udp_table,
1490 	.seq_fops	= &udp6_afinfo_seq_fops,
1491 	.seq_ops	= {
1492 		.show		= udp6_seq_show,
1493 	},
1494 };
1495 
1496 int __net_init udp6_proc_init(struct net *net)
1497 {
1498 	return udp_proc_register(net, &udp6_seq_afinfo);
1499 }
1500 
1501 void udp6_proc_exit(struct net *net) {
1502 	udp_proc_unregister(net, &udp6_seq_afinfo);
1503 }
1504 #endif /* CONFIG_PROC_FS */
1505 
1506 void udp_v6_clear_sk(struct sock *sk, int size)
1507 {
1508 	struct inet_sock *inet = inet_sk(sk);
1509 
1510 	/* we do not want to clear pinet6 field, because of RCU lookups */
1511 	sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1512 
1513 	size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1514 	memset(&inet->pinet6 + 1, 0, size);
1515 }
1516 
1517 /* ------------------------------------------------------------------------ */
1518 
1519 struct proto udpv6_prot = {
1520 	.name		   = "UDPv6",
1521 	.owner		   = THIS_MODULE,
1522 	.close		   = udp_lib_close,
1523 	.connect	   = ip6_datagram_connect,
1524 	.disconnect	   = udp_disconnect,
1525 	.ioctl		   = udp_ioctl,
1526 	.destroy	   = udpv6_destroy_sock,
1527 	.setsockopt	   = udpv6_setsockopt,
1528 	.getsockopt	   = udpv6_getsockopt,
1529 	.sendmsg	   = udpv6_sendmsg,
1530 	.recvmsg	   = udpv6_recvmsg,
1531 	.backlog_rcv	   = __udpv6_queue_rcv_skb,
1532 	.hash		   = udp_lib_hash,
1533 	.unhash		   = udp_lib_unhash,
1534 	.rehash		   = udp_v6_rehash,
1535 	.get_port	   = udp_v6_get_port,
1536 	.memory_allocated  = &udp_memory_allocated,
1537 	.sysctl_mem	   = sysctl_udp_mem,
1538 	.sysctl_wmem	   = &sysctl_udp_wmem_min,
1539 	.sysctl_rmem	   = &sysctl_udp_rmem_min,
1540 	.obj_size	   = sizeof(struct udp6_sock),
1541 	.slab_flags	   = SLAB_DESTROY_BY_RCU,
1542 	.h.udp_table	   = &udp_table,
1543 #ifdef CONFIG_COMPAT
1544 	.compat_setsockopt = compat_udpv6_setsockopt,
1545 	.compat_getsockopt = compat_udpv6_getsockopt,
1546 #endif
1547 	.clear_sk	   = udp_v6_clear_sk,
1548 };
1549 
1550 static struct inet_protosw udpv6_protosw = {
1551 	.type =      SOCK_DGRAM,
1552 	.protocol =  IPPROTO_UDP,
1553 	.prot =      &udpv6_prot,
1554 	.ops =       &inet6_dgram_ops,
1555 	.flags =     INET_PROTOSW_PERMANENT,
1556 };
1557 
1558 int __init udpv6_init(void)
1559 {
1560 	int ret;
1561 
1562 	ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1563 	if (ret)
1564 		goto out;
1565 
1566 	ret = inet6_register_protosw(&udpv6_protosw);
1567 	if (ret)
1568 		goto out_udpv6_protocol;
1569 out:
1570 	return ret;
1571 
1572 out_udpv6_protocol:
1573 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1574 	goto out;
1575 }
1576 
1577 void udpv6_exit(void)
1578 {
1579 	inet6_unregister_protosw(&udpv6_protosw);
1580 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1581 }
1582