xref: /openbmc/linux/net/ipv6/udp.c (revision 239480ab)
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 <linux/uaccess.h>
39 
40 #include <net/addrconf.h>
41 #include <net/ndisc.h>
42 #include <net/protocol.h>
43 #include <net/transp_v6.h>
44 #include <net/ip6_route.h>
45 #include <net/raw.h>
46 #include <net/tcp_states.h>
47 #include <net/ip6_checksum.h>
48 #include <net/xfrm.h>
49 #include <net/inet_hashtables.h>
50 #include <net/inet6_hashtables.h>
51 #include <net/busy_poll.h>
52 #include <net/sock_reuseport.h>
53 
54 #include <linux/proc_fs.h>
55 #include <linux/seq_file.h>
56 #include <trace/events/skb.h>
57 #include "udp_impl.h"
58 
59 static bool udp6_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
60 {
61 #if defined(CONFIG_NET_L3_MASTER_DEV)
62 	if (!net->ipv4.sysctl_udp_l3mdev_accept &&
63 	    skb && ipv6_l3mdev_skb(IP6CB(skb)->flags))
64 		return true;
65 #endif
66 	return false;
67 }
68 
69 static u32 udp6_ehashfn(const struct net *net,
70 			const struct in6_addr *laddr,
71 			const u16 lport,
72 			const struct in6_addr *faddr,
73 			const __be16 fport)
74 {
75 	static u32 udp6_ehash_secret __read_mostly;
76 	static u32 udp_ipv6_hash_secret __read_mostly;
77 
78 	u32 lhash, fhash;
79 
80 	net_get_random_once(&udp6_ehash_secret,
81 			    sizeof(udp6_ehash_secret));
82 	net_get_random_once(&udp_ipv6_hash_secret,
83 			    sizeof(udp_ipv6_hash_secret));
84 
85 	lhash = (__force u32)laddr->s6_addr32[3];
86 	fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
87 
88 	return __inet6_ehashfn(lhash, lport, fhash, fport,
89 			       udp_ipv6_hash_secret + net_hash_mix(net));
90 }
91 
92 static u32 udp6_portaddr_hash(const struct net *net,
93 			      const struct in6_addr *addr6,
94 			      unsigned int port)
95 {
96 	unsigned int hash, mix = net_hash_mix(net);
97 
98 	if (ipv6_addr_any(addr6))
99 		hash = jhash_1word(0, mix);
100 	else if (ipv6_addr_v4mapped(addr6))
101 		hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
102 	else
103 		hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
104 
105 	return hash ^ port;
106 }
107 
108 int udp_v6_get_port(struct sock *sk, unsigned short snum)
109 {
110 	unsigned int hash2_nulladdr =
111 		udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
112 	unsigned int hash2_partial =
113 		udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
114 
115 	/* precompute partial secondary hash */
116 	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
117 	return udp_lib_get_port(sk, snum, hash2_nulladdr);
118 }
119 
120 static void udp_v6_rehash(struct sock *sk)
121 {
122 	u16 new_hash = udp6_portaddr_hash(sock_net(sk),
123 					  &sk->sk_v6_rcv_saddr,
124 					  inet_sk(sk)->inet_num);
125 
126 	udp_lib_rehash(sk, new_hash);
127 }
128 
129 static int compute_score(struct sock *sk, struct net *net,
130 			 const struct in6_addr *saddr, __be16 sport,
131 			 const struct in6_addr *daddr, unsigned short hnum,
132 			 int dif, bool exact_dif)
133 {
134 	int score;
135 	struct inet_sock *inet;
136 
137 	if (!net_eq(sock_net(sk), net) ||
138 	    udp_sk(sk)->udp_port_hash != hnum ||
139 	    sk->sk_family != PF_INET6)
140 		return -1;
141 
142 	score = 0;
143 	inet = inet_sk(sk);
144 
145 	if (inet->inet_dport) {
146 		if (inet->inet_dport != sport)
147 			return -1;
148 		score++;
149 	}
150 
151 	if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
152 		if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
153 			return -1;
154 		score++;
155 	}
156 
157 	if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
158 		if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
159 			return -1;
160 		score++;
161 	}
162 
163 	if (sk->sk_bound_dev_if || exact_dif) {
164 		if (sk->sk_bound_dev_if != dif)
165 			return -1;
166 		score++;
167 	}
168 
169 	if (sk->sk_incoming_cpu == raw_smp_processor_id())
170 		score++;
171 
172 	return score;
173 }
174 
175 /* called with rcu_read_lock() */
176 static struct sock *udp6_lib_lookup2(struct net *net,
177 		const struct in6_addr *saddr, __be16 sport,
178 		const struct in6_addr *daddr, unsigned int hnum, int dif,
179 		bool exact_dif, struct udp_hslot *hslot2,
180 		struct sk_buff *skb)
181 {
182 	struct sock *sk, *result;
183 	int score, badness, matches = 0, reuseport = 0;
184 	u32 hash = 0;
185 
186 	result = NULL;
187 	badness = -1;
188 	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
189 		score = compute_score(sk, net, saddr, sport,
190 				      daddr, hnum, dif, exact_dif);
191 		if (score > badness) {
192 			reuseport = sk->sk_reuseport;
193 			if (reuseport) {
194 				hash = udp6_ehashfn(net, daddr, hnum,
195 						    saddr, sport);
196 
197 				result = reuseport_select_sock(sk, hash, skb,
198 							sizeof(struct udphdr));
199 				if (result)
200 					return result;
201 				matches = 1;
202 			}
203 			result = sk;
204 			badness = score;
205 		} else if (score == badness && reuseport) {
206 			matches++;
207 			if (reciprocal_scale(hash, matches) == 0)
208 				result = sk;
209 			hash = next_pseudo_random32(hash);
210 		}
211 	}
212 	return result;
213 }
214 
215 /* rcu_read_lock() must be held */
216 struct sock *__udp6_lib_lookup(struct net *net,
217 				      const struct in6_addr *saddr, __be16 sport,
218 				      const struct in6_addr *daddr, __be16 dport,
219 				      int dif, struct udp_table *udptable,
220 				      struct sk_buff *skb)
221 {
222 	struct sock *sk, *result;
223 	unsigned short hnum = ntohs(dport);
224 	unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
225 	struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
226 	bool exact_dif = udp6_lib_exact_dif_match(net, skb);
227 	int score, badness, matches = 0, reuseport = 0;
228 	u32 hash = 0;
229 
230 	if (hslot->count > 10) {
231 		hash2 = udp6_portaddr_hash(net, daddr, hnum);
232 		slot2 = hash2 & udptable->mask;
233 		hslot2 = &udptable->hash2[slot2];
234 		if (hslot->count < hslot2->count)
235 			goto begin;
236 
237 		result = udp6_lib_lookup2(net, saddr, sport,
238 					  daddr, hnum, dif, exact_dif,
239 					  hslot2, skb);
240 		if (!result) {
241 			unsigned int old_slot2 = slot2;
242 			hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
243 			slot2 = hash2 & udptable->mask;
244 			/* avoid searching the same slot again. */
245 			if (unlikely(slot2 == old_slot2))
246 				return result;
247 
248 			hslot2 = &udptable->hash2[slot2];
249 			if (hslot->count < hslot2->count)
250 				goto begin;
251 
252 			result = udp6_lib_lookup2(net, saddr, sport,
253 						  daddr, hnum, dif,
254 						  exact_dif, hslot2,
255 						  skb);
256 		}
257 		return result;
258 	}
259 begin:
260 	result = NULL;
261 	badness = -1;
262 	sk_for_each_rcu(sk, &hslot->head) {
263 		score = compute_score(sk, net, saddr, sport, daddr, hnum, dif,
264 				      exact_dif);
265 		if (score > badness) {
266 			reuseport = sk->sk_reuseport;
267 			if (reuseport) {
268 				hash = udp6_ehashfn(net, daddr, hnum,
269 						    saddr, sport);
270 				result = reuseport_select_sock(sk, hash, skb,
271 							sizeof(struct udphdr));
272 				if (result)
273 					return result;
274 				matches = 1;
275 			}
276 			result = sk;
277 			badness = score;
278 		} else if (score == badness && reuseport) {
279 			matches++;
280 			if (reciprocal_scale(hash, matches) == 0)
281 				result = sk;
282 			hash = next_pseudo_random32(hash);
283 		}
284 	}
285 	return result;
286 }
287 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
288 
289 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
290 					  __be16 sport, __be16 dport,
291 					  struct udp_table *udptable)
292 {
293 	const struct ipv6hdr *iph = ipv6_hdr(skb);
294 	struct sock *sk;
295 
296 	sk = skb_steal_sock(skb);
297 	if (unlikely(sk))
298 		return sk;
299 	return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
300 				 &iph->daddr, dport, inet6_iif(skb),
301 				 udptable, skb);
302 }
303 
304 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
305 				 __be16 sport, __be16 dport)
306 {
307 	const struct ipv6hdr *iph = ipv6_hdr(skb);
308 
309 	return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
310 				 &iph->daddr, dport, inet6_iif(skb),
311 				 &udp_table, skb);
312 }
313 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb);
314 
315 /* Must be called under rcu_read_lock().
316  * Does increment socket refcount.
317  */
318 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
319     IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
320     IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
321 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
322 			     const struct in6_addr *daddr, __be16 dport, int dif)
323 {
324 	struct sock *sk;
325 
326 	sk =  __udp6_lib_lookup(net, saddr, sport, daddr, dport,
327 				dif, &udp_table, NULL);
328 	if (sk && !atomic_inc_not_zero(&sk->sk_refcnt))
329 		sk = NULL;
330 	return sk;
331 }
332 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
333 #endif
334 
335 /*
336  *	This should be easy, if there is something there we
337  *	return it, otherwise we block.
338  */
339 
340 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
341 		  int noblock, int flags, int *addr_len)
342 {
343 	struct ipv6_pinfo *np = inet6_sk(sk);
344 	struct inet_sock *inet = inet_sk(sk);
345 	struct sk_buff *skb;
346 	unsigned int ulen, copied;
347 	int peeked, peeking, off;
348 	int err;
349 	int is_udplite = IS_UDPLITE(sk);
350 	bool checksum_valid = false;
351 	int is_udp4;
352 
353 	if (flags & MSG_ERRQUEUE)
354 		return ipv6_recv_error(sk, msg, len, addr_len);
355 
356 	if (np->rxpmtu && np->rxopt.bits.rxpmtu)
357 		return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
358 
359 try_again:
360 	peeking = off = sk_peek_offset(sk, flags);
361 	skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
362 	if (!skb)
363 		return err;
364 
365 	ulen = skb->len;
366 	copied = len;
367 	if (copied > ulen - off)
368 		copied = ulen - off;
369 	else if (copied < ulen)
370 		msg->msg_flags |= MSG_TRUNC;
371 
372 	is_udp4 = (skb->protocol == htons(ETH_P_IP));
373 
374 	/*
375 	 * If checksum is needed at all, try to do it while copying the
376 	 * data.  If the data is truncated, or if we only want a partial
377 	 * coverage checksum (UDP-Lite), do it before the copy.
378 	 */
379 
380 	if (copied < ulen || peeking ||
381 	    (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
382 		checksum_valid = !udp_lib_checksum_complete(skb);
383 		if (!checksum_valid)
384 			goto csum_copy_err;
385 	}
386 
387 	if (checksum_valid || skb_csum_unnecessary(skb))
388 		err = skb_copy_datagram_msg(skb, off, msg, copied);
389 	else {
390 		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
391 		if (err == -EINVAL)
392 			goto csum_copy_err;
393 	}
394 	if (unlikely(err)) {
395 		if (!peeked) {
396 			atomic_inc(&sk->sk_drops);
397 			if (is_udp4)
398 				UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
399 					      is_udplite);
400 			else
401 				UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
402 					       is_udplite);
403 		}
404 		kfree_skb(skb);
405 		return err;
406 	}
407 	if (!peeked) {
408 		if (is_udp4)
409 			UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
410 				      is_udplite);
411 		else
412 			UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
413 				       is_udplite);
414 	}
415 
416 	sock_recv_ts_and_drops(msg, sk, skb);
417 
418 	/* Copy the address. */
419 	if (msg->msg_name) {
420 		DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
421 		sin6->sin6_family = AF_INET6;
422 		sin6->sin6_port = udp_hdr(skb)->source;
423 		sin6->sin6_flowinfo = 0;
424 
425 		if (is_udp4) {
426 			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
427 					       &sin6->sin6_addr);
428 			sin6->sin6_scope_id = 0;
429 		} else {
430 			sin6->sin6_addr = ipv6_hdr(skb)->saddr;
431 			sin6->sin6_scope_id =
432 				ipv6_iface_scope_id(&sin6->sin6_addr,
433 						    inet6_iif(skb));
434 		}
435 		*addr_len = sizeof(*sin6);
436 	}
437 
438 	if (np->rxopt.all)
439 		ip6_datagram_recv_common_ctl(sk, msg, skb);
440 
441 	if (is_udp4) {
442 		if (inet->cmsg_flags)
443 			ip_cmsg_recv_offset(msg, sk, skb,
444 					    sizeof(struct udphdr), off);
445 	} else {
446 		if (np->rxopt.all)
447 			ip6_datagram_recv_specific_ctl(sk, msg, skb);
448 	}
449 
450 	err = copied;
451 	if (flags & MSG_TRUNC)
452 		err = ulen;
453 
454 	skb_consume_udp(sk, skb, peeking ? -err : err);
455 	return err;
456 
457 csum_copy_err:
458 	if (!__sk_queue_drop_skb(sk, skb, flags, udp_skb_destructor)) {
459 		if (is_udp4) {
460 			UDP_INC_STATS(sock_net(sk),
461 				      UDP_MIB_CSUMERRORS, is_udplite);
462 			UDP_INC_STATS(sock_net(sk),
463 				      UDP_MIB_INERRORS, is_udplite);
464 		} else {
465 			UDP6_INC_STATS(sock_net(sk),
466 				       UDP_MIB_CSUMERRORS, is_udplite);
467 			UDP6_INC_STATS(sock_net(sk),
468 				       UDP_MIB_INERRORS, is_udplite);
469 		}
470 	}
471 	kfree_skb(skb);
472 
473 	/* starting over for a new packet, but check if we need to yield */
474 	cond_resched();
475 	msg->msg_flags &= ~MSG_TRUNC;
476 	goto try_again;
477 }
478 
479 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
480 		    u8 type, u8 code, int offset, __be32 info,
481 		    struct udp_table *udptable)
482 {
483 	struct ipv6_pinfo *np;
484 	const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
485 	const struct in6_addr *saddr = &hdr->saddr;
486 	const struct in6_addr *daddr = &hdr->daddr;
487 	struct udphdr *uh = (struct udphdr *)(skb->data+offset);
488 	struct sock *sk;
489 	int harderr;
490 	int err;
491 	struct net *net = dev_net(skb->dev);
492 
493 	sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
494 			       inet6_iif(skb), udptable, skb);
495 	if (!sk) {
496 		__ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
497 				  ICMP6_MIB_INERRORS);
498 		return;
499 	}
500 
501 	harderr = icmpv6_err_convert(type, code, &err);
502 	np = inet6_sk(sk);
503 
504 	if (type == ICMPV6_PKT_TOOBIG) {
505 		if (!ip6_sk_accept_pmtu(sk))
506 			goto out;
507 		ip6_sk_update_pmtu(skb, sk, info);
508 		if (np->pmtudisc != IPV6_PMTUDISC_DONT)
509 			harderr = 1;
510 	}
511 	if (type == NDISC_REDIRECT) {
512 		ip6_sk_redirect(skb, sk);
513 		goto out;
514 	}
515 
516 	if (!np->recverr) {
517 		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
518 			goto out;
519 	} else {
520 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
521 	}
522 
523 	sk->sk_err = err;
524 	sk->sk_error_report(sk);
525 out:
526 	return;
527 }
528 
529 int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
530 {
531 	int rc;
532 
533 	if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
534 		sock_rps_save_rxhash(sk, skb);
535 		sk_mark_napi_id(sk, skb);
536 		sk_incoming_cpu_update(sk);
537 	} else {
538 		sk_mark_napi_id_once(sk, skb);
539 	}
540 
541 	rc = __udp_enqueue_schedule_skb(sk, skb);
542 	if (rc < 0) {
543 		int is_udplite = IS_UDPLITE(sk);
544 
545 		/* Note that an ENOMEM error is charged twice */
546 		if (rc == -ENOMEM)
547 			UDP6_INC_STATS(sock_net(sk),
548 					 UDP_MIB_RCVBUFERRORS, is_udplite);
549 		UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
550 		kfree_skb(skb);
551 		return -1;
552 	}
553 
554 	return 0;
555 }
556 
557 static __inline__ void udpv6_err(struct sk_buff *skb,
558 				 struct inet6_skb_parm *opt, u8 type,
559 				 u8 code, int offset, __be32 info)
560 {
561 	__udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
562 }
563 
564 static struct static_key udpv6_encap_needed __read_mostly;
565 void udpv6_encap_enable(void)
566 {
567 	if (!static_key_enabled(&udpv6_encap_needed))
568 		static_key_slow_inc(&udpv6_encap_needed);
569 }
570 EXPORT_SYMBOL(udpv6_encap_enable);
571 
572 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
573 {
574 	struct udp_sock *up = udp_sk(sk);
575 	int is_udplite = IS_UDPLITE(sk);
576 
577 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
578 		goto drop;
579 
580 	if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
581 		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
582 
583 		/*
584 		 * This is an encapsulation socket so pass the skb to
585 		 * the socket's udp_encap_rcv() hook. Otherwise, just
586 		 * fall through and pass this up the UDP socket.
587 		 * up->encap_rcv() returns the following value:
588 		 * =0 if skb was successfully passed to the encap
589 		 *    handler or was discarded by it.
590 		 * >0 if skb should be passed on to UDP.
591 		 * <0 if skb should be resubmitted as proto -N
592 		 */
593 
594 		/* if we're overly short, let UDP handle it */
595 		encap_rcv = ACCESS_ONCE(up->encap_rcv);
596 		if (encap_rcv) {
597 			int ret;
598 
599 			/* Verify checksum before giving to encap */
600 			if (udp_lib_checksum_complete(skb))
601 				goto csum_error;
602 
603 			ret = encap_rcv(sk, skb);
604 			if (ret <= 0) {
605 				__UDP_INC_STATS(sock_net(sk),
606 						UDP_MIB_INDATAGRAMS,
607 						is_udplite);
608 				return -ret;
609 			}
610 		}
611 
612 		/* FALLTHROUGH -- it's a UDP Packet */
613 	}
614 
615 	/*
616 	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
617 	 */
618 	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
619 
620 		if (up->pcrlen == 0) {          /* full coverage was set  */
621 			net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
622 					    UDP_SKB_CB(skb)->cscov, skb->len);
623 			goto drop;
624 		}
625 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
626 			net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
627 					    UDP_SKB_CB(skb)->cscov, up->pcrlen);
628 			goto drop;
629 		}
630 	}
631 
632 	if (rcu_access_pointer(sk->sk_filter) &&
633 	    udp_lib_checksum_complete(skb))
634 		goto csum_error;
635 
636 	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
637 		goto drop;
638 
639 	udp_csum_pull_header(skb);
640 
641 	skb_dst_drop(skb);
642 
643 	return __udpv6_queue_rcv_skb(sk, skb);
644 
645 csum_error:
646 	__UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
647 drop:
648 	__UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
649 	atomic_inc(&sk->sk_drops);
650 	kfree_skb(skb);
651 	return -1;
652 }
653 
654 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
655 				   __be16 loc_port, const struct in6_addr *loc_addr,
656 				   __be16 rmt_port, const struct in6_addr *rmt_addr,
657 				   int dif, unsigned short hnum)
658 {
659 	struct inet_sock *inet = inet_sk(sk);
660 
661 	if (!net_eq(sock_net(sk), net))
662 		return false;
663 
664 	if (udp_sk(sk)->udp_port_hash != hnum ||
665 	    sk->sk_family != PF_INET6 ||
666 	    (inet->inet_dport && inet->inet_dport != rmt_port) ||
667 	    (!ipv6_addr_any(&sk->sk_v6_daddr) &&
668 		    !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
669 	    (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
670 	    (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
671 		    !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
672 		return false;
673 	if (!inet6_mc_check(sk, loc_addr, rmt_addr))
674 		return false;
675 	return true;
676 }
677 
678 static void udp6_csum_zero_error(struct sk_buff *skb)
679 {
680 	/* RFC 2460 section 8.1 says that we SHOULD log
681 	 * this error. Well, it is reasonable.
682 	 */
683 	net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
684 			    &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
685 			    &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
686 }
687 
688 /*
689  * Note: called only from the BH handler context,
690  * so we don't need to lock the hashes.
691  */
692 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
693 		const struct in6_addr *saddr, const struct in6_addr *daddr,
694 		struct udp_table *udptable, int proto)
695 {
696 	struct sock *sk, *first = NULL;
697 	const struct udphdr *uh = udp_hdr(skb);
698 	unsigned short hnum = ntohs(uh->dest);
699 	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
700 	unsigned int offset = offsetof(typeof(*sk), sk_node);
701 	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
702 	int dif = inet6_iif(skb);
703 	struct hlist_node *node;
704 	struct sk_buff *nskb;
705 
706 	if (use_hash2) {
707 		hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
708 			    udptable->mask;
709 		hash2 = udp6_portaddr_hash(net, daddr, hnum) & udptable->mask;
710 start_lookup:
711 		hslot = &udptable->hash2[hash2];
712 		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
713 	}
714 
715 	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
716 		if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
717 					    uh->source, saddr, dif, hnum))
718 			continue;
719 		/* If zero checksum and no_check is not on for
720 		 * the socket then skip it.
721 		 */
722 		if (!uh->check && !udp_sk(sk)->no_check6_rx)
723 			continue;
724 		if (!first) {
725 			first = sk;
726 			continue;
727 		}
728 		nskb = skb_clone(skb, GFP_ATOMIC);
729 		if (unlikely(!nskb)) {
730 			atomic_inc(&sk->sk_drops);
731 			__UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
732 					 IS_UDPLITE(sk));
733 			__UDP6_INC_STATS(net, UDP_MIB_INERRORS,
734 					 IS_UDPLITE(sk));
735 			continue;
736 		}
737 
738 		if (udpv6_queue_rcv_skb(sk, nskb) > 0)
739 			consume_skb(nskb);
740 	}
741 
742 	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
743 	if (use_hash2 && hash2 != hash2_any) {
744 		hash2 = hash2_any;
745 		goto start_lookup;
746 	}
747 
748 	if (first) {
749 		if (udpv6_queue_rcv_skb(first, skb) > 0)
750 			consume_skb(skb);
751 	} else {
752 		kfree_skb(skb);
753 		__UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
754 				 proto == IPPROTO_UDPLITE);
755 	}
756 	return 0;
757 }
758 
759 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
760 		   int proto)
761 {
762 	const struct in6_addr *saddr, *daddr;
763 	struct net *net = dev_net(skb->dev);
764 	struct udphdr *uh;
765 	struct sock *sk;
766 	u32 ulen = 0;
767 
768 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
769 		goto discard;
770 
771 	saddr = &ipv6_hdr(skb)->saddr;
772 	daddr = &ipv6_hdr(skb)->daddr;
773 	uh = udp_hdr(skb);
774 
775 	ulen = ntohs(uh->len);
776 	if (ulen > skb->len)
777 		goto short_packet;
778 
779 	if (proto == IPPROTO_UDP) {
780 		/* UDP validates ulen. */
781 
782 		/* Check for jumbo payload */
783 		if (ulen == 0)
784 			ulen = skb->len;
785 
786 		if (ulen < sizeof(*uh))
787 			goto short_packet;
788 
789 		if (ulen < skb->len) {
790 			if (pskb_trim_rcsum(skb, ulen))
791 				goto short_packet;
792 			saddr = &ipv6_hdr(skb)->saddr;
793 			daddr = &ipv6_hdr(skb)->daddr;
794 			uh = udp_hdr(skb);
795 		}
796 	}
797 
798 	if (udp6_csum_init(skb, uh, proto))
799 		goto csum_error;
800 
801 	/*
802 	 *	Multicast receive code
803 	 */
804 	if (ipv6_addr_is_multicast(daddr))
805 		return __udp6_lib_mcast_deliver(net, skb,
806 				saddr, daddr, udptable, proto);
807 
808 	/* Unicast */
809 
810 	/*
811 	 * check socket cache ... must talk to Alan about his plans
812 	 * for sock caches... i'll skip this for now.
813 	 */
814 	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
815 	if (sk) {
816 		int ret;
817 
818 		if (!uh->check && !udp_sk(sk)->no_check6_rx) {
819 			udp6_csum_zero_error(skb);
820 			goto csum_error;
821 		}
822 
823 		if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
824 			skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
825 						 ip6_compute_pseudo);
826 
827 		ret = udpv6_queue_rcv_skb(sk, skb);
828 
829 		/* a return value > 0 means to resubmit the input */
830 		if (ret > 0)
831 			return ret;
832 
833 		return 0;
834 	}
835 
836 	if (!uh->check) {
837 		udp6_csum_zero_error(skb);
838 		goto csum_error;
839 	}
840 
841 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
842 		goto discard;
843 
844 	if (udp_lib_checksum_complete(skb))
845 		goto csum_error;
846 
847 	__UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
848 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
849 
850 	kfree_skb(skb);
851 	return 0;
852 
853 short_packet:
854 	net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
855 			    proto == IPPROTO_UDPLITE ? "-Lite" : "",
856 			    saddr, ntohs(uh->source),
857 			    ulen, skb->len,
858 			    daddr, ntohs(uh->dest));
859 	goto discard;
860 csum_error:
861 	__UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
862 discard:
863 	__UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
864 	kfree_skb(skb);
865 	return 0;
866 }
867 
868 
869 static struct sock *__udp6_lib_demux_lookup(struct net *net,
870 			__be16 loc_port, const struct in6_addr *loc_addr,
871 			__be16 rmt_port, const struct in6_addr *rmt_addr,
872 			int dif)
873 {
874 	unsigned short hnum = ntohs(loc_port);
875 	unsigned int hash2 = udp6_portaddr_hash(net, loc_addr, hnum);
876 	unsigned int slot2 = hash2 & udp_table.mask;
877 	struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
878 	const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
879 	struct sock *sk;
880 
881 	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
882 		if (INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif))
883 			return sk;
884 		/* Only check first socket in chain */
885 		break;
886 	}
887 	return NULL;
888 }
889 
890 static void udp_v6_early_demux(struct sk_buff *skb)
891 {
892 	struct net *net = dev_net(skb->dev);
893 	const struct udphdr *uh;
894 	struct sock *sk;
895 	struct dst_entry *dst;
896 	int dif = skb->dev->ifindex;
897 
898 	if (!pskb_may_pull(skb, skb_transport_offset(skb) +
899 	    sizeof(struct udphdr)))
900 		return;
901 
902 	uh = udp_hdr(skb);
903 
904 	if (skb->pkt_type == PACKET_HOST)
905 		sk = __udp6_lib_demux_lookup(net, uh->dest,
906 					     &ipv6_hdr(skb)->daddr,
907 					     uh->source, &ipv6_hdr(skb)->saddr,
908 					     dif);
909 	else
910 		return;
911 
912 	if (!sk || !atomic_inc_not_zero_hint(&sk->sk_refcnt, 2))
913 		return;
914 
915 	skb->sk = sk;
916 	skb->destructor = sock_efree;
917 	dst = READ_ONCE(sk->sk_rx_dst);
918 
919 	if (dst)
920 		dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
921 	if (dst) {
922 		if (dst->flags & DST_NOCACHE) {
923 			if (likely(atomic_inc_not_zero(&dst->__refcnt)))
924 				skb_dst_set(skb, dst);
925 		} else {
926 			skb_dst_set_noref(skb, dst);
927 		}
928 	}
929 }
930 
931 static __inline__ int udpv6_rcv(struct sk_buff *skb)
932 {
933 	return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
934 }
935 
936 /*
937  * Throw away all pending data and cancel the corking. Socket is locked.
938  */
939 static void udp_v6_flush_pending_frames(struct sock *sk)
940 {
941 	struct udp_sock *up = udp_sk(sk);
942 
943 	if (up->pending == AF_INET)
944 		udp_flush_pending_frames(sk);
945 	else if (up->pending) {
946 		up->len = 0;
947 		up->pending = 0;
948 		ip6_flush_pending_frames(sk);
949 	}
950 }
951 
952 /**
953  *	udp6_hwcsum_outgoing  -  handle outgoing HW checksumming
954  *	@sk:	socket we are sending on
955  *	@skb:	sk_buff containing the filled-in UDP header
956  *		(checksum field must be zeroed out)
957  */
958 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
959 				 const struct in6_addr *saddr,
960 				 const struct in6_addr *daddr, int len)
961 {
962 	unsigned int offset;
963 	struct udphdr *uh = udp_hdr(skb);
964 	struct sk_buff *frags = skb_shinfo(skb)->frag_list;
965 	__wsum csum = 0;
966 
967 	if (!frags) {
968 		/* Only one fragment on the socket.  */
969 		skb->csum_start = skb_transport_header(skb) - skb->head;
970 		skb->csum_offset = offsetof(struct udphdr, check);
971 		uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
972 	} else {
973 		/*
974 		 * HW-checksum won't work as there are two or more
975 		 * fragments on the socket so that all csums of sk_buffs
976 		 * should be together
977 		 */
978 		offset = skb_transport_offset(skb);
979 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
980 
981 		skb->ip_summed = CHECKSUM_NONE;
982 
983 		do {
984 			csum = csum_add(csum, frags->csum);
985 		} while ((frags = frags->next));
986 
987 		uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
988 					    csum);
989 		if (uh->check == 0)
990 			uh->check = CSUM_MANGLED_0;
991 	}
992 }
993 
994 /*
995  *	Sending
996  */
997 
998 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
999 {
1000 	struct sock *sk = skb->sk;
1001 	struct udphdr *uh;
1002 	int err = 0;
1003 	int is_udplite = IS_UDPLITE(sk);
1004 	__wsum csum = 0;
1005 	int offset = skb_transport_offset(skb);
1006 	int len = skb->len - offset;
1007 
1008 	/*
1009 	 * Create a UDP header
1010 	 */
1011 	uh = udp_hdr(skb);
1012 	uh->source = fl6->fl6_sport;
1013 	uh->dest = fl6->fl6_dport;
1014 	uh->len = htons(len);
1015 	uh->check = 0;
1016 
1017 	if (is_udplite)
1018 		csum = udplite_csum(skb);
1019 	else if (udp_sk(sk)->no_check6_tx) {   /* UDP csum disabled */
1020 		skb->ip_summed = CHECKSUM_NONE;
1021 		goto send;
1022 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1023 		udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1024 		goto send;
1025 	} else
1026 		csum = udp_csum(skb);
1027 
1028 	/* add protocol-dependent pseudo-header */
1029 	uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1030 				    len, fl6->flowi6_proto, csum);
1031 	if (uh->check == 0)
1032 		uh->check = CSUM_MANGLED_0;
1033 
1034 send:
1035 	err = ip6_send_skb(skb);
1036 	if (err) {
1037 		if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1038 			UDP6_INC_STATS(sock_net(sk),
1039 				       UDP_MIB_SNDBUFERRORS, is_udplite);
1040 			err = 0;
1041 		}
1042 	} else {
1043 		UDP6_INC_STATS(sock_net(sk),
1044 			       UDP_MIB_OUTDATAGRAMS, is_udplite);
1045 	}
1046 	return err;
1047 }
1048 
1049 static int udp_v6_push_pending_frames(struct sock *sk)
1050 {
1051 	struct sk_buff *skb;
1052 	struct udp_sock  *up = udp_sk(sk);
1053 	struct flowi6 fl6;
1054 	int err = 0;
1055 
1056 	if (up->pending == AF_INET)
1057 		return udp_push_pending_frames(sk);
1058 
1059 	/* ip6_finish_skb will release the cork, so make a copy of
1060 	 * fl6 here.
1061 	 */
1062 	fl6 = inet_sk(sk)->cork.fl.u.ip6;
1063 
1064 	skb = ip6_finish_skb(sk);
1065 	if (!skb)
1066 		goto out;
1067 
1068 	err = udp_v6_send_skb(skb, &fl6);
1069 
1070 out:
1071 	up->len = 0;
1072 	up->pending = 0;
1073 	return err;
1074 }
1075 
1076 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1077 {
1078 	struct ipv6_txoptions opt_space;
1079 	struct udp_sock *up = udp_sk(sk);
1080 	struct inet_sock *inet = inet_sk(sk);
1081 	struct ipv6_pinfo *np = inet6_sk(sk);
1082 	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1083 	struct in6_addr *daddr, *final_p, final;
1084 	struct ipv6_txoptions *opt = NULL;
1085 	struct ipv6_txoptions *opt_to_free = NULL;
1086 	struct ip6_flowlabel *flowlabel = NULL;
1087 	struct flowi6 fl6;
1088 	struct dst_entry *dst;
1089 	struct ipcm6_cookie ipc6;
1090 	int addr_len = msg->msg_namelen;
1091 	int ulen = len;
1092 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1093 	int err;
1094 	int connected = 0;
1095 	int is_udplite = IS_UDPLITE(sk);
1096 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1097 	struct sockcm_cookie sockc;
1098 
1099 	ipc6.hlimit = -1;
1100 	ipc6.tclass = -1;
1101 	ipc6.dontfrag = -1;
1102 	sockc.tsflags = sk->sk_tsflags;
1103 
1104 	/* destination address check */
1105 	if (sin6) {
1106 		if (addr_len < offsetof(struct sockaddr, sa_data))
1107 			return -EINVAL;
1108 
1109 		switch (sin6->sin6_family) {
1110 		case AF_INET6:
1111 			if (addr_len < SIN6_LEN_RFC2133)
1112 				return -EINVAL;
1113 			daddr = &sin6->sin6_addr;
1114 			if (ipv6_addr_any(daddr) &&
1115 			    ipv6_addr_v4mapped(&np->saddr))
1116 				ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1117 						       daddr);
1118 			break;
1119 		case AF_INET:
1120 			goto do_udp_sendmsg;
1121 		case AF_UNSPEC:
1122 			msg->msg_name = sin6 = NULL;
1123 			msg->msg_namelen = addr_len = 0;
1124 			daddr = NULL;
1125 			break;
1126 		default:
1127 			return -EINVAL;
1128 		}
1129 	} else if (!up->pending) {
1130 		if (sk->sk_state != TCP_ESTABLISHED)
1131 			return -EDESTADDRREQ;
1132 		daddr = &sk->sk_v6_daddr;
1133 	} else
1134 		daddr = NULL;
1135 
1136 	if (daddr) {
1137 		if (ipv6_addr_v4mapped(daddr)) {
1138 			struct sockaddr_in sin;
1139 			sin.sin_family = AF_INET;
1140 			sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1141 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
1142 			msg->msg_name = &sin;
1143 			msg->msg_namelen = sizeof(sin);
1144 do_udp_sendmsg:
1145 			if (__ipv6_only_sock(sk))
1146 				return -ENETUNREACH;
1147 			return udp_sendmsg(sk, msg, len);
1148 		}
1149 	}
1150 
1151 	if (up->pending == AF_INET)
1152 		return udp_sendmsg(sk, msg, len);
1153 
1154 	/* Rough check on arithmetic overflow,
1155 	   better check is made in ip6_append_data().
1156 	   */
1157 	if (len > INT_MAX - sizeof(struct udphdr))
1158 		return -EMSGSIZE;
1159 
1160 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
1161 	if (up->pending) {
1162 		/*
1163 		 * There are pending frames.
1164 		 * The socket lock must be held while it's corked.
1165 		 */
1166 		lock_sock(sk);
1167 		if (likely(up->pending)) {
1168 			if (unlikely(up->pending != AF_INET6)) {
1169 				release_sock(sk);
1170 				return -EAFNOSUPPORT;
1171 			}
1172 			dst = NULL;
1173 			goto do_append_data;
1174 		}
1175 		release_sock(sk);
1176 	}
1177 	ulen += sizeof(struct udphdr);
1178 
1179 	memset(&fl6, 0, sizeof(fl6));
1180 
1181 	if (sin6) {
1182 		if (sin6->sin6_port == 0)
1183 			return -EINVAL;
1184 
1185 		fl6.fl6_dport = sin6->sin6_port;
1186 		daddr = &sin6->sin6_addr;
1187 
1188 		if (np->sndflow) {
1189 			fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1190 			if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1191 				flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1192 				if (!flowlabel)
1193 					return -EINVAL;
1194 			}
1195 		}
1196 
1197 		/*
1198 		 * Otherwise it will be difficult to maintain
1199 		 * sk->sk_dst_cache.
1200 		 */
1201 		if (sk->sk_state == TCP_ESTABLISHED &&
1202 		    ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1203 			daddr = &sk->sk_v6_daddr;
1204 
1205 		if (addr_len >= sizeof(struct sockaddr_in6) &&
1206 		    sin6->sin6_scope_id &&
1207 		    __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1208 			fl6.flowi6_oif = sin6->sin6_scope_id;
1209 	} else {
1210 		if (sk->sk_state != TCP_ESTABLISHED)
1211 			return -EDESTADDRREQ;
1212 
1213 		fl6.fl6_dport = inet->inet_dport;
1214 		daddr = &sk->sk_v6_daddr;
1215 		fl6.flowlabel = np->flow_label;
1216 		connected = 1;
1217 	}
1218 
1219 	if (!fl6.flowi6_oif)
1220 		fl6.flowi6_oif = sk->sk_bound_dev_if;
1221 
1222 	if (!fl6.flowi6_oif)
1223 		fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1224 
1225 	fl6.flowi6_mark = sk->sk_mark;
1226 	fl6.flowi6_uid = sk->sk_uid;
1227 
1228 	if (msg->msg_controllen) {
1229 		opt = &opt_space;
1230 		memset(opt, 0, sizeof(struct ipv6_txoptions));
1231 		opt->tot_len = sizeof(*opt);
1232 		ipc6.opt = opt;
1233 
1234 		err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6, &sockc);
1235 		if (err < 0) {
1236 			fl6_sock_release(flowlabel);
1237 			return err;
1238 		}
1239 		if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1240 			flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1241 			if (!flowlabel)
1242 				return -EINVAL;
1243 		}
1244 		if (!(opt->opt_nflen|opt->opt_flen))
1245 			opt = NULL;
1246 		connected = 0;
1247 	}
1248 	if (!opt) {
1249 		opt = txopt_get(np);
1250 		opt_to_free = opt;
1251 	}
1252 	if (flowlabel)
1253 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
1254 	opt = ipv6_fixup_options(&opt_space, opt);
1255 	ipc6.opt = opt;
1256 
1257 	fl6.flowi6_proto = sk->sk_protocol;
1258 	if (!ipv6_addr_any(daddr))
1259 		fl6.daddr = *daddr;
1260 	else
1261 		fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1262 	if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1263 		fl6.saddr = np->saddr;
1264 	fl6.fl6_sport = inet->inet_sport;
1265 
1266 	final_p = fl6_update_dst(&fl6, opt, &final);
1267 	if (final_p)
1268 		connected = 0;
1269 
1270 	if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1271 		fl6.flowi6_oif = np->mcast_oif;
1272 		connected = 0;
1273 	} else if (!fl6.flowi6_oif)
1274 		fl6.flowi6_oif = np->ucast_oif;
1275 
1276 	security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1277 
1278 	if (ipc6.tclass < 0)
1279 		ipc6.tclass = np->tclass;
1280 
1281 	fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1282 
1283 	dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1284 	if (IS_ERR(dst)) {
1285 		err = PTR_ERR(dst);
1286 		dst = NULL;
1287 		goto out;
1288 	}
1289 
1290 	if (ipc6.hlimit < 0)
1291 		ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1292 
1293 	if (msg->msg_flags&MSG_CONFIRM)
1294 		goto do_confirm;
1295 back_from_confirm:
1296 
1297 	/* Lockless fast path for the non-corking case */
1298 	if (!corkreq) {
1299 		struct sk_buff *skb;
1300 
1301 		skb = ip6_make_skb(sk, getfrag, msg, ulen,
1302 				   sizeof(struct udphdr), &ipc6,
1303 				   &fl6, (struct rt6_info *)dst,
1304 				   msg->msg_flags, &sockc);
1305 		err = PTR_ERR(skb);
1306 		if (!IS_ERR_OR_NULL(skb))
1307 			err = udp_v6_send_skb(skb, &fl6);
1308 		goto release_dst;
1309 	}
1310 
1311 	lock_sock(sk);
1312 	if (unlikely(up->pending)) {
1313 		/* The socket is already corked while preparing it. */
1314 		/* ... which is an evident application bug. --ANK */
1315 		release_sock(sk);
1316 
1317 		net_dbg_ratelimited("udp cork app bug 2\n");
1318 		err = -EINVAL;
1319 		goto out;
1320 	}
1321 
1322 	up->pending = AF_INET6;
1323 
1324 do_append_data:
1325 	if (ipc6.dontfrag < 0)
1326 		ipc6.dontfrag = np->dontfrag;
1327 	up->len += ulen;
1328 	err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1329 			      &ipc6, &fl6, (struct rt6_info *)dst,
1330 			      corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, &sockc);
1331 	if (err)
1332 		udp_v6_flush_pending_frames(sk);
1333 	else if (!corkreq)
1334 		err = udp_v6_push_pending_frames(sk);
1335 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1336 		up->pending = 0;
1337 
1338 	if (err > 0)
1339 		err = np->recverr ? net_xmit_errno(err) : 0;
1340 	release_sock(sk);
1341 
1342 release_dst:
1343 	if (dst) {
1344 		if (connected) {
1345 			ip6_dst_store(sk, dst,
1346 				      ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1347 				      &sk->sk_v6_daddr : NULL,
1348 #ifdef CONFIG_IPV6_SUBTREES
1349 				      ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1350 				      &np->saddr :
1351 #endif
1352 				      NULL);
1353 		} else {
1354 			dst_release(dst);
1355 		}
1356 		dst = NULL;
1357 	}
1358 
1359 out:
1360 	dst_release(dst);
1361 	fl6_sock_release(flowlabel);
1362 	txopt_put(opt_to_free);
1363 	if (!err)
1364 		return len;
1365 	/*
1366 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1367 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1368 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1369 	 * things).  We could add another new stat but at least for now that
1370 	 * seems like overkill.
1371 	 */
1372 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1373 		UDP6_INC_STATS(sock_net(sk),
1374 			       UDP_MIB_SNDBUFERRORS, is_udplite);
1375 	}
1376 	return err;
1377 
1378 do_confirm:
1379 	if (msg->msg_flags & MSG_PROBE)
1380 		dst_confirm_neigh(dst, &fl6.daddr);
1381 	if (!(msg->msg_flags&MSG_PROBE) || len)
1382 		goto back_from_confirm;
1383 	err = 0;
1384 	goto out;
1385 }
1386 
1387 void udpv6_destroy_sock(struct sock *sk)
1388 {
1389 	struct udp_sock *up = udp_sk(sk);
1390 	lock_sock(sk);
1391 	udp_v6_flush_pending_frames(sk);
1392 	release_sock(sk);
1393 
1394 	if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1395 		void (*encap_destroy)(struct sock *sk);
1396 		encap_destroy = ACCESS_ONCE(up->encap_destroy);
1397 		if (encap_destroy)
1398 			encap_destroy(sk);
1399 	}
1400 
1401 	inet6_destroy_sock(sk);
1402 }
1403 
1404 /*
1405  *	Socket option code for UDP
1406  */
1407 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1408 		     char __user *optval, unsigned int optlen)
1409 {
1410 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1411 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1412 					  udp_v6_push_pending_frames);
1413 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
1414 }
1415 
1416 #ifdef CONFIG_COMPAT
1417 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1418 			    char __user *optval, unsigned int optlen)
1419 {
1420 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1421 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1422 					  udp_v6_push_pending_frames);
1423 	return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1424 }
1425 #endif
1426 
1427 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1428 		     char __user *optval, int __user *optlen)
1429 {
1430 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1431 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1432 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
1433 }
1434 
1435 #ifdef CONFIG_COMPAT
1436 int compat_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 compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1442 }
1443 #endif
1444 
1445 static struct inet6_protocol udpv6_protocol = {
1446 	.early_demux	=	udp_v6_early_demux,
1447 	.early_demux_handler =  udp_v6_early_demux,
1448 	.handler	=	udpv6_rcv,
1449 	.err_handler	=	udpv6_err,
1450 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1451 };
1452 
1453 /* ------------------------------------------------------------------------ */
1454 #ifdef CONFIG_PROC_FS
1455 int udp6_seq_show(struct seq_file *seq, void *v)
1456 {
1457 	if (v == SEQ_START_TOKEN) {
1458 		seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1459 	} else {
1460 		int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1461 		struct inet_sock *inet = inet_sk(v);
1462 		__u16 srcp = ntohs(inet->inet_sport);
1463 		__u16 destp = ntohs(inet->inet_dport);
1464 		ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1465 	}
1466 	return 0;
1467 }
1468 
1469 static const struct file_operations udp6_afinfo_seq_fops = {
1470 	.owner    = THIS_MODULE,
1471 	.open     = udp_seq_open,
1472 	.read     = seq_read,
1473 	.llseek   = seq_lseek,
1474 	.release  = seq_release_net
1475 };
1476 
1477 static struct udp_seq_afinfo udp6_seq_afinfo = {
1478 	.name		= "udp6",
1479 	.family		= AF_INET6,
1480 	.udp_table	= &udp_table,
1481 	.seq_fops	= &udp6_afinfo_seq_fops,
1482 	.seq_ops	= {
1483 		.show		= udp6_seq_show,
1484 	},
1485 };
1486 
1487 int __net_init udp6_proc_init(struct net *net)
1488 {
1489 	return udp_proc_register(net, &udp6_seq_afinfo);
1490 }
1491 
1492 void udp6_proc_exit(struct net *net)
1493 {
1494 	udp_proc_unregister(net, &udp6_seq_afinfo);
1495 }
1496 #endif /* CONFIG_PROC_FS */
1497 
1498 /* ------------------------------------------------------------------------ */
1499 
1500 struct proto udpv6_prot = {
1501 	.name		   = "UDPv6",
1502 	.owner		   = THIS_MODULE,
1503 	.close		   = udp_lib_close,
1504 	.connect	   = ip6_datagram_connect,
1505 	.disconnect	   = udp_disconnect,
1506 	.ioctl		   = udp_ioctl,
1507 	.init		   = udp_init_sock,
1508 	.destroy	   = udpv6_destroy_sock,
1509 	.setsockopt	   = udpv6_setsockopt,
1510 	.getsockopt	   = udpv6_getsockopt,
1511 	.sendmsg	   = udpv6_sendmsg,
1512 	.recvmsg	   = udpv6_recvmsg,
1513 	.release_cb	   = ip6_datagram_release_cb,
1514 	.hash		   = udp_lib_hash,
1515 	.unhash		   = udp_lib_unhash,
1516 	.rehash		   = udp_v6_rehash,
1517 	.get_port	   = udp_v6_get_port,
1518 	.memory_allocated  = &udp_memory_allocated,
1519 	.sysctl_mem	   = sysctl_udp_mem,
1520 	.sysctl_wmem	   = &sysctl_udp_wmem_min,
1521 	.sysctl_rmem	   = &sysctl_udp_rmem_min,
1522 	.obj_size	   = sizeof(struct udp6_sock),
1523 	.h.udp_table	   = &udp_table,
1524 #ifdef CONFIG_COMPAT
1525 	.compat_setsockopt = compat_udpv6_setsockopt,
1526 	.compat_getsockopt = compat_udpv6_getsockopt,
1527 #endif
1528 	.diag_destroy      = udp_abort,
1529 };
1530 
1531 static struct inet_protosw udpv6_protosw = {
1532 	.type =      SOCK_DGRAM,
1533 	.protocol =  IPPROTO_UDP,
1534 	.prot =      &udpv6_prot,
1535 	.ops =       &inet6_dgram_ops,
1536 	.flags =     INET_PROTOSW_PERMANENT,
1537 };
1538 
1539 int __init udpv6_init(void)
1540 {
1541 	int ret;
1542 
1543 	ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1544 	if (ret)
1545 		goto out;
1546 
1547 	ret = inet6_register_protosw(&udpv6_protosw);
1548 	if (ret)
1549 		goto out_udpv6_protocol;
1550 out:
1551 	return ret;
1552 
1553 out_udpv6_protocol:
1554 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1555 	goto out;
1556 }
1557 
1558 void udpv6_exit(void)
1559 {
1560 	inet6_unregister_protosw(&udpv6_protosw);
1561 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1562 }
1563