xref: /openbmc/linux/net/ipv6/udp.c (revision 4bce6fce)
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 		return false;
736 	if (!inet6_mc_check(sk, loc_addr, rmt_addr))
737 		return false;
738 	return true;
739 }
740 
741 static void flush_stack(struct sock **stack, unsigned int count,
742 			struct sk_buff *skb, unsigned int final)
743 {
744 	struct sk_buff *skb1 = NULL;
745 	struct sock *sk;
746 	unsigned int i;
747 
748 	for (i = 0; i < count; i++) {
749 		sk = stack[i];
750 		if (likely(!skb1))
751 			skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
752 		if (!skb1) {
753 			atomic_inc(&sk->sk_drops);
754 			UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
755 					  IS_UDPLITE(sk));
756 			UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
757 					  IS_UDPLITE(sk));
758 		}
759 
760 		if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
761 			skb1 = NULL;
762 		sock_put(sk);
763 	}
764 	if (unlikely(skb1))
765 		kfree_skb(skb1);
766 }
767 
768 static void udp6_csum_zero_error(struct sk_buff *skb)
769 {
770 	/* RFC 2460 section 8.1 says that we SHOULD log
771 	 * this error. Well, it is reasonable.
772 	 */
773 	net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
774 			    &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
775 			    &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
776 }
777 
778 /*
779  * Note: called only from the BH handler context,
780  * so we don't need to lock the hashes.
781  */
782 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
783 		const struct in6_addr *saddr, const struct in6_addr *daddr,
784 		struct udp_table *udptable, int proto)
785 {
786 	struct sock *sk, *stack[256 / sizeof(struct sock *)];
787 	const struct udphdr *uh = udp_hdr(skb);
788 	struct hlist_nulls_node *node;
789 	unsigned short hnum = ntohs(uh->dest);
790 	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
791 	int dif = inet6_iif(skb);
792 	unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
793 	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
794 	bool inner_flushed = false;
795 
796 	if (use_hash2) {
797 		hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
798 			    udp_table.mask;
799 		hash2 = udp6_portaddr_hash(net, daddr, hnum) & udp_table.mask;
800 start_lookup:
801 		hslot = &udp_table.hash2[hash2];
802 		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
803 	}
804 
805 	spin_lock(&hslot->lock);
806 	sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
807 		if (__udp_v6_is_mcast_sock(net, sk,
808 					   uh->dest, daddr,
809 					   uh->source, saddr,
810 					   dif, hnum) &&
811 		    /* If zero checksum and no_check is not on for
812 		     * the socket then skip it.
813 		     */
814 		    (uh->check || udp_sk(sk)->no_check6_rx)) {
815 			if (unlikely(count == ARRAY_SIZE(stack))) {
816 				flush_stack(stack, count, skb, ~0);
817 				inner_flushed = true;
818 				count = 0;
819 			}
820 			stack[count++] = sk;
821 			sock_hold(sk);
822 		}
823 	}
824 
825 	spin_unlock(&hslot->lock);
826 
827 	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
828 	if (use_hash2 && hash2 != hash2_any) {
829 		hash2 = hash2_any;
830 		goto start_lookup;
831 	}
832 
833 	if (count) {
834 		flush_stack(stack, count, skb, count - 1);
835 	} else {
836 		if (!inner_flushed)
837 			UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
838 					 proto == IPPROTO_UDPLITE);
839 		consume_skb(skb);
840 	}
841 	return 0;
842 }
843 
844 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
845 		   int proto)
846 {
847 	struct net *net = dev_net(skb->dev);
848 	struct sock *sk;
849 	struct udphdr *uh;
850 	const struct in6_addr *saddr, *daddr;
851 	u32 ulen = 0;
852 
853 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
854 		goto discard;
855 
856 	saddr = &ipv6_hdr(skb)->saddr;
857 	daddr = &ipv6_hdr(skb)->daddr;
858 	uh = udp_hdr(skb);
859 
860 	ulen = ntohs(uh->len);
861 	if (ulen > skb->len)
862 		goto short_packet;
863 
864 	if (proto == IPPROTO_UDP) {
865 		/* UDP validates ulen. */
866 
867 		/* Check for jumbo payload */
868 		if (ulen == 0)
869 			ulen = skb->len;
870 
871 		if (ulen < sizeof(*uh))
872 			goto short_packet;
873 
874 		if (ulen < skb->len) {
875 			if (pskb_trim_rcsum(skb, ulen))
876 				goto short_packet;
877 			saddr = &ipv6_hdr(skb)->saddr;
878 			daddr = &ipv6_hdr(skb)->daddr;
879 			uh = udp_hdr(skb);
880 		}
881 	}
882 
883 	if (udp6_csum_init(skb, uh, proto))
884 		goto csum_error;
885 
886 	/*
887 	 *	Multicast receive code
888 	 */
889 	if (ipv6_addr_is_multicast(daddr))
890 		return __udp6_lib_mcast_deliver(net, skb,
891 				saddr, daddr, udptable, proto);
892 
893 	/* Unicast */
894 
895 	/*
896 	 * check socket cache ... must talk to Alan about his plans
897 	 * for sock caches... i'll skip this for now.
898 	 */
899 	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
900 	if (sk) {
901 		int ret;
902 
903 		if (!uh->check && !udp_sk(sk)->no_check6_rx) {
904 			sock_put(sk);
905 			udp6_csum_zero_error(skb);
906 			goto csum_error;
907 		}
908 
909 		if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
910 			skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
911 						 ip6_compute_pseudo);
912 
913 		ret = udpv6_queue_rcv_skb(sk, skb);
914 		sock_put(sk);
915 
916 		/* a return value > 0 means to resubmit the input, but
917 		 * it wants the return to be -protocol, or 0
918 		 */
919 		if (ret > 0)
920 			return -ret;
921 
922 		return 0;
923 	}
924 
925 	if (!uh->check) {
926 		udp6_csum_zero_error(skb);
927 		goto csum_error;
928 	}
929 
930 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
931 		goto discard;
932 
933 	if (udp_lib_checksum_complete(skb))
934 		goto csum_error;
935 
936 	UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
937 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
938 
939 	kfree_skb(skb);
940 	return 0;
941 
942 short_packet:
943 	net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
944 			    proto == IPPROTO_UDPLITE ? "-Lite" : "",
945 			    saddr, ntohs(uh->source),
946 			    ulen, skb->len,
947 			    daddr, ntohs(uh->dest));
948 	goto discard;
949 csum_error:
950 	UDP6_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
951 discard:
952 	UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
953 	kfree_skb(skb);
954 	return 0;
955 }
956 
957 static __inline__ int udpv6_rcv(struct sk_buff *skb)
958 {
959 	return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
960 }
961 
962 /*
963  * Throw away all pending data and cancel the corking. Socket is locked.
964  */
965 static void udp_v6_flush_pending_frames(struct sock *sk)
966 {
967 	struct udp_sock *up = udp_sk(sk);
968 
969 	if (up->pending == AF_INET)
970 		udp_flush_pending_frames(sk);
971 	else if (up->pending) {
972 		up->len = 0;
973 		up->pending = 0;
974 		ip6_flush_pending_frames(sk);
975 	}
976 }
977 
978 /**
979  *	udp6_hwcsum_outgoing  -  handle outgoing HW checksumming
980  *	@sk:	socket we are sending on
981  *	@skb:	sk_buff containing the filled-in UDP header
982  *		(checksum field must be zeroed out)
983  */
984 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
985 				 const struct in6_addr *saddr,
986 				 const struct in6_addr *daddr, int len)
987 {
988 	unsigned int offset;
989 	struct udphdr *uh = udp_hdr(skb);
990 	struct sk_buff *frags = skb_shinfo(skb)->frag_list;
991 	__wsum csum = 0;
992 
993 	if (!frags) {
994 		/* Only one fragment on the socket.  */
995 		skb->csum_start = skb_transport_header(skb) - skb->head;
996 		skb->csum_offset = offsetof(struct udphdr, check);
997 		uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
998 	} else {
999 		/*
1000 		 * HW-checksum won't work as there are two or more
1001 		 * fragments on the socket so that all csums of sk_buffs
1002 		 * should be together
1003 		 */
1004 		offset = skb_transport_offset(skb);
1005 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1006 
1007 		skb->ip_summed = CHECKSUM_NONE;
1008 
1009 		do {
1010 			csum = csum_add(csum, frags->csum);
1011 		} while ((frags = frags->next));
1012 
1013 		uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1014 					    csum);
1015 		if (uh->check == 0)
1016 			uh->check = CSUM_MANGLED_0;
1017 	}
1018 }
1019 
1020 /*
1021  *	Sending
1022  */
1023 
1024 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
1025 {
1026 	struct sock *sk = skb->sk;
1027 	struct udphdr *uh;
1028 	int err = 0;
1029 	int is_udplite = IS_UDPLITE(sk);
1030 	__wsum csum = 0;
1031 	int offset = skb_transport_offset(skb);
1032 	int len = skb->len - offset;
1033 
1034 	/*
1035 	 * Create a UDP header
1036 	 */
1037 	uh = udp_hdr(skb);
1038 	uh->source = fl6->fl6_sport;
1039 	uh->dest = fl6->fl6_dport;
1040 	uh->len = htons(len);
1041 	uh->check = 0;
1042 
1043 	if (is_udplite)
1044 		csum = udplite_csum(skb);
1045 	else if (udp_sk(sk)->no_check6_tx) {   /* UDP csum disabled */
1046 		skb->ip_summed = CHECKSUM_NONE;
1047 		goto send;
1048 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1049 		udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1050 		goto send;
1051 	} else
1052 		csum = udp_csum(skb);
1053 
1054 	/* add protocol-dependent pseudo-header */
1055 	uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1056 				    len, fl6->flowi6_proto, csum);
1057 	if (uh->check == 0)
1058 		uh->check = CSUM_MANGLED_0;
1059 
1060 send:
1061 	err = ip6_send_skb(skb);
1062 	if (err) {
1063 		if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1064 			UDP6_INC_STATS_USER(sock_net(sk),
1065 					    UDP_MIB_SNDBUFERRORS, is_udplite);
1066 			err = 0;
1067 		}
1068 	} else
1069 		UDP6_INC_STATS_USER(sock_net(sk),
1070 				    UDP_MIB_OUTDATAGRAMS, is_udplite);
1071 	return err;
1072 }
1073 
1074 static int udp_v6_push_pending_frames(struct sock *sk)
1075 {
1076 	struct sk_buff *skb;
1077 	struct udp_sock  *up = udp_sk(sk);
1078 	struct flowi6 fl6;
1079 	int err = 0;
1080 
1081 	if (up->pending == AF_INET)
1082 		return udp_push_pending_frames(sk);
1083 
1084 	/* ip6_finish_skb will release the cork, so make a copy of
1085 	 * fl6 here.
1086 	 */
1087 	fl6 = inet_sk(sk)->cork.fl.u.ip6;
1088 
1089 	skb = ip6_finish_skb(sk);
1090 	if (!skb)
1091 		goto out;
1092 
1093 	err = udp_v6_send_skb(skb, &fl6);
1094 
1095 out:
1096 	up->len = 0;
1097 	up->pending = 0;
1098 	return err;
1099 }
1100 
1101 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1102 {
1103 	struct ipv6_txoptions opt_space;
1104 	struct udp_sock *up = udp_sk(sk);
1105 	struct inet_sock *inet = inet_sk(sk);
1106 	struct ipv6_pinfo *np = inet6_sk(sk);
1107 	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1108 	struct in6_addr *daddr, *final_p, final;
1109 	struct ipv6_txoptions *opt = NULL;
1110 	struct ip6_flowlabel *flowlabel = NULL;
1111 	struct flowi6 fl6;
1112 	struct dst_entry *dst;
1113 	int addr_len = msg->msg_namelen;
1114 	int ulen = len;
1115 	int hlimit = -1;
1116 	int tclass = -1;
1117 	int dontfrag = -1;
1118 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1119 	int err;
1120 	int connected = 0;
1121 	int is_udplite = IS_UDPLITE(sk);
1122 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1123 
1124 	/* destination address check */
1125 	if (sin6) {
1126 		if (addr_len < offsetof(struct sockaddr, sa_data))
1127 			return -EINVAL;
1128 
1129 		switch (sin6->sin6_family) {
1130 		case AF_INET6:
1131 			if (addr_len < SIN6_LEN_RFC2133)
1132 				return -EINVAL;
1133 			daddr = &sin6->sin6_addr;
1134 			break;
1135 		case AF_INET:
1136 			goto do_udp_sendmsg;
1137 		case AF_UNSPEC:
1138 			msg->msg_name = sin6 = NULL;
1139 			msg->msg_namelen = addr_len = 0;
1140 			daddr = NULL;
1141 			break;
1142 		default:
1143 			return -EINVAL;
1144 		}
1145 	} else if (!up->pending) {
1146 		if (sk->sk_state != TCP_ESTABLISHED)
1147 			return -EDESTADDRREQ;
1148 		daddr = &sk->sk_v6_daddr;
1149 	} else
1150 		daddr = NULL;
1151 
1152 	if (daddr) {
1153 		if (ipv6_addr_v4mapped(daddr)) {
1154 			struct sockaddr_in sin;
1155 			sin.sin_family = AF_INET;
1156 			sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1157 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
1158 			msg->msg_name = &sin;
1159 			msg->msg_namelen = sizeof(sin);
1160 do_udp_sendmsg:
1161 			if (__ipv6_only_sock(sk))
1162 				return -ENETUNREACH;
1163 			return udp_sendmsg(sk, msg, len);
1164 		}
1165 	}
1166 
1167 	if (up->pending == AF_INET)
1168 		return udp_sendmsg(sk, msg, len);
1169 
1170 	/* Rough check on arithmetic overflow,
1171 	   better check is made in ip6_append_data().
1172 	   */
1173 	if (len > INT_MAX - sizeof(struct udphdr))
1174 		return -EMSGSIZE;
1175 
1176 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
1177 	if (up->pending) {
1178 		/*
1179 		 * There are pending frames.
1180 		 * The socket lock must be held while it's corked.
1181 		 */
1182 		lock_sock(sk);
1183 		if (likely(up->pending)) {
1184 			if (unlikely(up->pending != AF_INET6)) {
1185 				release_sock(sk);
1186 				return -EAFNOSUPPORT;
1187 			}
1188 			dst = NULL;
1189 			goto do_append_data;
1190 		}
1191 		release_sock(sk);
1192 	}
1193 	ulen += sizeof(struct udphdr);
1194 
1195 	memset(&fl6, 0, sizeof(fl6));
1196 
1197 	if (sin6) {
1198 		if (sin6->sin6_port == 0)
1199 			return -EINVAL;
1200 
1201 		fl6.fl6_dport = sin6->sin6_port;
1202 		daddr = &sin6->sin6_addr;
1203 
1204 		if (np->sndflow) {
1205 			fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1206 			if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1207 				flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1208 				if (!flowlabel)
1209 					return -EINVAL;
1210 			}
1211 		}
1212 
1213 		/*
1214 		 * Otherwise it will be difficult to maintain
1215 		 * sk->sk_dst_cache.
1216 		 */
1217 		if (sk->sk_state == TCP_ESTABLISHED &&
1218 		    ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1219 			daddr = &sk->sk_v6_daddr;
1220 
1221 		if (addr_len >= sizeof(struct sockaddr_in6) &&
1222 		    sin6->sin6_scope_id &&
1223 		    __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1224 			fl6.flowi6_oif = sin6->sin6_scope_id;
1225 	} else {
1226 		if (sk->sk_state != TCP_ESTABLISHED)
1227 			return -EDESTADDRREQ;
1228 
1229 		fl6.fl6_dport = inet->inet_dport;
1230 		daddr = &sk->sk_v6_daddr;
1231 		fl6.flowlabel = np->flow_label;
1232 		connected = 1;
1233 	}
1234 
1235 	if (!fl6.flowi6_oif)
1236 		fl6.flowi6_oif = sk->sk_bound_dev_if;
1237 
1238 	if (!fl6.flowi6_oif)
1239 		fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1240 
1241 	fl6.flowi6_mark = sk->sk_mark;
1242 
1243 	if (msg->msg_controllen) {
1244 		opt = &opt_space;
1245 		memset(opt, 0, sizeof(struct ipv6_txoptions));
1246 		opt->tot_len = sizeof(*opt);
1247 
1248 		err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1249 					    &hlimit, &tclass, &dontfrag);
1250 		if (err < 0) {
1251 			fl6_sock_release(flowlabel);
1252 			return err;
1253 		}
1254 		if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1255 			flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1256 			if (!flowlabel)
1257 				return -EINVAL;
1258 		}
1259 		if (!(opt->opt_nflen|opt->opt_flen))
1260 			opt = NULL;
1261 		connected = 0;
1262 	}
1263 	if (!opt)
1264 		opt = np->opt;
1265 	if (flowlabel)
1266 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
1267 	opt = ipv6_fixup_options(&opt_space, opt);
1268 
1269 	fl6.flowi6_proto = sk->sk_protocol;
1270 	if (!ipv6_addr_any(daddr))
1271 		fl6.daddr = *daddr;
1272 	else
1273 		fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1274 	if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1275 		fl6.saddr = np->saddr;
1276 	fl6.fl6_sport = inet->inet_sport;
1277 
1278 	final_p = fl6_update_dst(&fl6, opt, &final);
1279 	if (final_p)
1280 		connected = 0;
1281 
1282 	if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1283 		fl6.flowi6_oif = np->mcast_oif;
1284 		connected = 0;
1285 	} else if (!fl6.flowi6_oif)
1286 		fl6.flowi6_oif = np->ucast_oif;
1287 
1288 	security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1289 
1290 	dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1291 	if (IS_ERR(dst)) {
1292 		err = PTR_ERR(dst);
1293 		dst = NULL;
1294 		goto out;
1295 	}
1296 
1297 	if (hlimit < 0)
1298 		hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1299 
1300 	if (tclass < 0)
1301 		tclass = np->tclass;
1302 
1303 	if (msg->msg_flags&MSG_CONFIRM)
1304 		goto do_confirm;
1305 back_from_confirm:
1306 
1307 	/* Lockless fast path for the non-corking case */
1308 	if (!corkreq) {
1309 		struct sk_buff *skb;
1310 
1311 		skb = ip6_make_skb(sk, getfrag, msg, ulen,
1312 				   sizeof(struct udphdr), hlimit, tclass, opt,
1313 				   &fl6, (struct rt6_info *)dst,
1314 				   msg->msg_flags, dontfrag);
1315 		err = PTR_ERR(skb);
1316 		if (!IS_ERR_OR_NULL(skb))
1317 			err = udp_v6_send_skb(skb, &fl6);
1318 		goto release_dst;
1319 	}
1320 
1321 	lock_sock(sk);
1322 	if (unlikely(up->pending)) {
1323 		/* The socket is already corked while preparing it. */
1324 		/* ... which is an evident application bug. --ANK */
1325 		release_sock(sk);
1326 
1327 		net_dbg_ratelimited("udp cork app bug 2\n");
1328 		err = -EINVAL;
1329 		goto out;
1330 	}
1331 
1332 	up->pending = AF_INET6;
1333 
1334 do_append_data:
1335 	if (dontfrag < 0)
1336 		dontfrag = np->dontfrag;
1337 	up->len += ulen;
1338 	err = ip6_append_data(sk, getfrag, msg, ulen,
1339 		sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1340 		(struct rt6_info *)dst,
1341 		corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1342 	if (err)
1343 		udp_v6_flush_pending_frames(sk);
1344 	else if (!corkreq)
1345 		err = udp_v6_push_pending_frames(sk);
1346 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1347 		up->pending = 0;
1348 
1349 	if (err > 0)
1350 		err = np->recverr ? net_xmit_errno(err) : 0;
1351 	release_sock(sk);
1352 
1353 release_dst:
1354 	if (dst) {
1355 		if (connected) {
1356 			ip6_dst_store(sk, dst,
1357 				      ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1358 				      &sk->sk_v6_daddr : NULL,
1359 #ifdef CONFIG_IPV6_SUBTREES
1360 				      ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1361 				      &np->saddr :
1362 #endif
1363 				      NULL);
1364 		} else {
1365 			dst_release(dst);
1366 		}
1367 		dst = NULL;
1368 	}
1369 
1370 out:
1371 	dst_release(dst);
1372 	fl6_sock_release(flowlabel);
1373 	if (!err)
1374 		return len;
1375 	/*
1376 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1377 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1378 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1379 	 * things).  We could add another new stat but at least for now that
1380 	 * seems like overkill.
1381 	 */
1382 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1383 		UDP6_INC_STATS_USER(sock_net(sk),
1384 				UDP_MIB_SNDBUFERRORS, is_udplite);
1385 	}
1386 	return err;
1387 
1388 do_confirm:
1389 	dst_confirm(dst);
1390 	if (!(msg->msg_flags&MSG_PROBE) || len)
1391 		goto back_from_confirm;
1392 	err = 0;
1393 	goto out;
1394 }
1395 
1396 void udpv6_destroy_sock(struct sock *sk)
1397 {
1398 	struct udp_sock *up = udp_sk(sk);
1399 	lock_sock(sk);
1400 	udp_v6_flush_pending_frames(sk);
1401 	release_sock(sk);
1402 
1403 	if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1404 		void (*encap_destroy)(struct sock *sk);
1405 		encap_destroy = ACCESS_ONCE(up->encap_destroy);
1406 		if (encap_destroy)
1407 			encap_destroy(sk);
1408 	}
1409 
1410 	inet6_destroy_sock(sk);
1411 }
1412 
1413 /*
1414  *	Socket option code for UDP
1415  */
1416 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1417 		     char __user *optval, unsigned int optlen)
1418 {
1419 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1420 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1421 					  udp_v6_push_pending_frames);
1422 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
1423 }
1424 
1425 #ifdef CONFIG_COMPAT
1426 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1427 			    char __user *optval, unsigned int optlen)
1428 {
1429 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1430 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1431 					  udp_v6_push_pending_frames);
1432 	return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1433 }
1434 #endif
1435 
1436 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1437 		     char __user *optval, int __user *optlen)
1438 {
1439 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1440 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1441 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
1442 }
1443 
1444 #ifdef CONFIG_COMPAT
1445 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1446 			    char __user *optval, int __user *optlen)
1447 {
1448 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1449 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1450 	return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1451 }
1452 #endif
1453 
1454 static const struct inet6_protocol udpv6_protocol = {
1455 	.handler	=	udpv6_rcv,
1456 	.err_handler	=	udpv6_err,
1457 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1458 };
1459 
1460 /* ------------------------------------------------------------------------ */
1461 #ifdef CONFIG_PROC_FS
1462 int udp6_seq_show(struct seq_file *seq, void *v)
1463 {
1464 	if (v == SEQ_START_TOKEN) {
1465 		seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1466 	} else {
1467 		int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1468 		struct inet_sock *inet = inet_sk(v);
1469 		__u16 srcp = ntohs(inet->inet_sport);
1470 		__u16 destp = ntohs(inet->inet_dport);
1471 		ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1472 	}
1473 	return 0;
1474 }
1475 
1476 static const struct file_operations udp6_afinfo_seq_fops = {
1477 	.owner    = THIS_MODULE,
1478 	.open     = udp_seq_open,
1479 	.read     = seq_read,
1480 	.llseek   = seq_lseek,
1481 	.release  = seq_release_net
1482 };
1483 
1484 static struct udp_seq_afinfo udp6_seq_afinfo = {
1485 	.name		= "udp6",
1486 	.family		= AF_INET6,
1487 	.udp_table	= &udp_table,
1488 	.seq_fops	= &udp6_afinfo_seq_fops,
1489 	.seq_ops	= {
1490 		.show		= udp6_seq_show,
1491 	},
1492 };
1493 
1494 int __net_init udp6_proc_init(struct net *net)
1495 {
1496 	return udp_proc_register(net, &udp6_seq_afinfo);
1497 }
1498 
1499 void udp6_proc_exit(struct net *net) {
1500 	udp_proc_unregister(net, &udp6_seq_afinfo);
1501 }
1502 #endif /* CONFIG_PROC_FS */
1503 
1504 void udp_v6_clear_sk(struct sock *sk, int size)
1505 {
1506 	struct inet_sock *inet = inet_sk(sk);
1507 
1508 	/* we do not want to clear pinet6 field, because of RCU lookups */
1509 	sk_prot_clear_portaddr_nulls(sk, offsetof(struct inet_sock, pinet6));
1510 
1511 	size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1512 	memset(&inet->pinet6 + 1, 0, size);
1513 }
1514 
1515 /* ------------------------------------------------------------------------ */
1516 
1517 struct proto udpv6_prot = {
1518 	.name		   = "UDPv6",
1519 	.owner		   = THIS_MODULE,
1520 	.close		   = udp_lib_close,
1521 	.connect	   = ip6_datagram_connect,
1522 	.disconnect	   = udp_disconnect,
1523 	.ioctl		   = udp_ioctl,
1524 	.destroy	   = udpv6_destroy_sock,
1525 	.setsockopt	   = udpv6_setsockopt,
1526 	.getsockopt	   = udpv6_getsockopt,
1527 	.sendmsg	   = udpv6_sendmsg,
1528 	.recvmsg	   = udpv6_recvmsg,
1529 	.backlog_rcv	   = __udpv6_queue_rcv_skb,
1530 	.hash		   = udp_lib_hash,
1531 	.unhash		   = udp_lib_unhash,
1532 	.rehash		   = udp_v6_rehash,
1533 	.get_port	   = udp_v6_get_port,
1534 	.memory_allocated  = &udp_memory_allocated,
1535 	.sysctl_mem	   = sysctl_udp_mem,
1536 	.sysctl_wmem	   = &sysctl_udp_wmem_min,
1537 	.sysctl_rmem	   = &sysctl_udp_rmem_min,
1538 	.obj_size	   = sizeof(struct udp6_sock),
1539 	.slab_flags	   = SLAB_DESTROY_BY_RCU,
1540 	.h.udp_table	   = &udp_table,
1541 #ifdef CONFIG_COMPAT
1542 	.compat_setsockopt = compat_udpv6_setsockopt,
1543 	.compat_getsockopt = compat_udpv6_getsockopt,
1544 #endif
1545 	.clear_sk	   = udp_v6_clear_sk,
1546 };
1547 
1548 static struct inet_protosw udpv6_protosw = {
1549 	.type =      SOCK_DGRAM,
1550 	.protocol =  IPPROTO_UDP,
1551 	.prot =      &udpv6_prot,
1552 	.ops =       &inet6_dgram_ops,
1553 	.flags =     INET_PROTOSW_PERMANENT,
1554 };
1555 
1556 int __init udpv6_init(void)
1557 {
1558 	int ret;
1559 
1560 	ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1561 	if (ret)
1562 		goto out;
1563 
1564 	ret = inet6_register_protosw(&udpv6_protosw);
1565 	if (ret)
1566 		goto out_udpv6_protocol;
1567 out:
1568 	return ret;
1569 
1570 out_udpv6_protocol:
1571 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1572 	goto out;
1573 }
1574 
1575 void udpv6_exit(void)
1576 {
1577 	inet6_unregister_protosw(&udpv6_protosw);
1578 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1579 }
1580