xref: /openbmc/linux/net/ipv6/udp.c (revision 5d0e4d78)
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 && !refcount_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 = udp_skb_len(skb);
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_skb_csum_unnecessary(skb) ||
383 				!__udp_lib_checksum_complete(skb);
384 		if (!checksum_valid)
385 			goto csum_copy_err;
386 	}
387 
388 	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
389 		if (udp_skb_is_linear(skb))
390 			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
391 		else
392 			err = skb_copy_datagram_msg(skb, off, msg, copied);
393 	} else {
394 		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
395 		if (err == -EINVAL)
396 			goto csum_copy_err;
397 	}
398 	if (unlikely(err)) {
399 		if (!peeked) {
400 			atomic_inc(&sk->sk_drops);
401 			if (is_udp4)
402 				UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
403 					      is_udplite);
404 			else
405 				UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
406 					       is_udplite);
407 		}
408 		kfree_skb(skb);
409 		return err;
410 	}
411 	if (!peeked) {
412 		if (is_udp4)
413 			UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
414 				      is_udplite);
415 		else
416 			UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
417 				       is_udplite);
418 	}
419 
420 	sock_recv_ts_and_drops(msg, sk, skb);
421 
422 	/* Copy the address. */
423 	if (msg->msg_name) {
424 		DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
425 		sin6->sin6_family = AF_INET6;
426 		sin6->sin6_port = udp_hdr(skb)->source;
427 		sin6->sin6_flowinfo = 0;
428 
429 		if (is_udp4) {
430 			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
431 					       &sin6->sin6_addr);
432 			sin6->sin6_scope_id = 0;
433 		} else {
434 			sin6->sin6_addr = ipv6_hdr(skb)->saddr;
435 			sin6->sin6_scope_id =
436 				ipv6_iface_scope_id(&sin6->sin6_addr,
437 						    inet6_iif(skb));
438 		}
439 		*addr_len = sizeof(*sin6);
440 	}
441 
442 	if (np->rxopt.all)
443 		ip6_datagram_recv_common_ctl(sk, msg, skb);
444 
445 	if (is_udp4) {
446 		if (inet->cmsg_flags)
447 			ip_cmsg_recv_offset(msg, sk, skb,
448 					    sizeof(struct udphdr), off);
449 	} else {
450 		if (np->rxopt.all)
451 			ip6_datagram_recv_specific_ctl(sk, msg, skb);
452 	}
453 
454 	err = copied;
455 	if (flags & MSG_TRUNC)
456 		err = ulen;
457 
458 	skb_consume_udp(sk, skb, peeking ? -err : err);
459 	return err;
460 
461 csum_copy_err:
462 	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
463 				 udp_skb_destructor)) {
464 		if (is_udp4) {
465 			UDP_INC_STATS(sock_net(sk),
466 				      UDP_MIB_CSUMERRORS, is_udplite);
467 			UDP_INC_STATS(sock_net(sk),
468 				      UDP_MIB_INERRORS, is_udplite);
469 		} else {
470 			UDP6_INC_STATS(sock_net(sk),
471 				       UDP_MIB_CSUMERRORS, is_udplite);
472 			UDP6_INC_STATS(sock_net(sk),
473 				       UDP_MIB_INERRORS, is_udplite);
474 		}
475 	}
476 	kfree_skb(skb);
477 
478 	/* starting over for a new packet, but check if we need to yield */
479 	cond_resched();
480 	msg->msg_flags &= ~MSG_TRUNC;
481 	goto try_again;
482 }
483 
484 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
485 		    u8 type, u8 code, int offset, __be32 info,
486 		    struct udp_table *udptable)
487 {
488 	struct ipv6_pinfo *np;
489 	const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
490 	const struct in6_addr *saddr = &hdr->saddr;
491 	const struct in6_addr *daddr = &hdr->daddr;
492 	struct udphdr *uh = (struct udphdr *)(skb->data+offset);
493 	struct sock *sk;
494 	int harderr;
495 	int err;
496 	struct net *net = dev_net(skb->dev);
497 
498 	sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
499 			       inet6_iif(skb), udptable, skb);
500 	if (!sk) {
501 		__ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
502 				  ICMP6_MIB_INERRORS);
503 		return;
504 	}
505 
506 	harderr = icmpv6_err_convert(type, code, &err);
507 	np = inet6_sk(sk);
508 
509 	if (type == ICMPV6_PKT_TOOBIG) {
510 		if (!ip6_sk_accept_pmtu(sk))
511 			goto out;
512 		ip6_sk_update_pmtu(skb, sk, info);
513 		if (np->pmtudisc != IPV6_PMTUDISC_DONT)
514 			harderr = 1;
515 	}
516 	if (type == NDISC_REDIRECT) {
517 		ip6_sk_redirect(skb, sk);
518 		goto out;
519 	}
520 
521 	if (!np->recverr) {
522 		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
523 			goto out;
524 	} else {
525 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
526 	}
527 
528 	sk->sk_err = err;
529 	sk->sk_error_report(sk);
530 out:
531 	return;
532 }
533 
534 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
535 {
536 	int rc;
537 
538 	if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
539 		sock_rps_save_rxhash(sk, skb);
540 		sk_mark_napi_id(sk, skb);
541 		sk_incoming_cpu_update(sk);
542 	} else {
543 		sk_mark_napi_id_once(sk, skb);
544 	}
545 
546 	rc = __udp_enqueue_schedule_skb(sk, skb);
547 	if (rc < 0) {
548 		int is_udplite = IS_UDPLITE(sk);
549 
550 		/* Note that an ENOMEM error is charged twice */
551 		if (rc == -ENOMEM)
552 			UDP6_INC_STATS(sock_net(sk),
553 					 UDP_MIB_RCVBUFERRORS, is_udplite);
554 		UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
555 		kfree_skb(skb);
556 		return -1;
557 	}
558 
559 	return 0;
560 }
561 
562 static __inline__ void udpv6_err(struct sk_buff *skb,
563 				 struct inet6_skb_parm *opt, u8 type,
564 				 u8 code, int offset, __be32 info)
565 {
566 	__udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
567 }
568 
569 static struct static_key udpv6_encap_needed __read_mostly;
570 void udpv6_encap_enable(void)
571 {
572 	if (!static_key_enabled(&udpv6_encap_needed))
573 		static_key_slow_inc(&udpv6_encap_needed);
574 }
575 EXPORT_SYMBOL(udpv6_encap_enable);
576 
577 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
578 {
579 	struct udp_sock *up = udp_sk(sk);
580 	int is_udplite = IS_UDPLITE(sk);
581 
582 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
583 		goto drop;
584 
585 	if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
586 		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
587 
588 		/*
589 		 * This is an encapsulation socket so pass the skb to
590 		 * the socket's udp_encap_rcv() hook. Otherwise, just
591 		 * fall through and pass this up the UDP socket.
592 		 * up->encap_rcv() returns the following value:
593 		 * =0 if skb was successfully passed to the encap
594 		 *    handler or was discarded by it.
595 		 * >0 if skb should be passed on to UDP.
596 		 * <0 if skb should be resubmitted as proto -N
597 		 */
598 
599 		/* if we're overly short, let UDP handle it */
600 		encap_rcv = ACCESS_ONCE(up->encap_rcv);
601 		if (encap_rcv) {
602 			int ret;
603 
604 			/* Verify checksum before giving to encap */
605 			if (udp_lib_checksum_complete(skb))
606 				goto csum_error;
607 
608 			ret = encap_rcv(sk, skb);
609 			if (ret <= 0) {
610 				__UDP_INC_STATS(sock_net(sk),
611 						UDP_MIB_INDATAGRAMS,
612 						is_udplite);
613 				return -ret;
614 			}
615 		}
616 
617 		/* FALLTHROUGH -- it's a UDP Packet */
618 	}
619 
620 	/*
621 	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
622 	 */
623 	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
624 
625 		if (up->pcrlen == 0) {          /* full coverage was set  */
626 			net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
627 					    UDP_SKB_CB(skb)->cscov, skb->len);
628 			goto drop;
629 		}
630 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
631 			net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
632 					    UDP_SKB_CB(skb)->cscov, up->pcrlen);
633 			goto drop;
634 		}
635 	}
636 
637 	prefetch(&sk->sk_rmem_alloc);
638 	if (rcu_access_pointer(sk->sk_filter) &&
639 	    udp_lib_checksum_complete(skb))
640 		goto csum_error;
641 
642 	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
643 		goto drop;
644 
645 	udp_csum_pull_header(skb);
646 
647 	skb_dst_drop(skb);
648 
649 	return __udpv6_queue_rcv_skb(sk, skb);
650 
651 csum_error:
652 	__UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
653 drop:
654 	__UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
655 	atomic_inc(&sk->sk_drops);
656 	kfree_skb(skb);
657 	return -1;
658 }
659 
660 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
661 				   __be16 loc_port, const struct in6_addr *loc_addr,
662 				   __be16 rmt_port, const struct in6_addr *rmt_addr,
663 				   int dif, unsigned short hnum)
664 {
665 	struct inet_sock *inet = inet_sk(sk);
666 
667 	if (!net_eq(sock_net(sk), net))
668 		return false;
669 
670 	if (udp_sk(sk)->udp_port_hash != hnum ||
671 	    sk->sk_family != PF_INET6 ||
672 	    (inet->inet_dport && inet->inet_dport != rmt_port) ||
673 	    (!ipv6_addr_any(&sk->sk_v6_daddr) &&
674 		    !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
675 	    (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
676 	    (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
677 		    !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
678 		return false;
679 	if (!inet6_mc_check(sk, loc_addr, rmt_addr))
680 		return false;
681 	return true;
682 }
683 
684 static void udp6_csum_zero_error(struct sk_buff *skb)
685 {
686 	/* RFC 2460 section 8.1 says that we SHOULD log
687 	 * this error. Well, it is reasonable.
688 	 */
689 	net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
690 			    &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
691 			    &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
692 }
693 
694 /*
695  * Note: called only from the BH handler context,
696  * so we don't need to lock the hashes.
697  */
698 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
699 		const struct in6_addr *saddr, const struct in6_addr *daddr,
700 		struct udp_table *udptable, int proto)
701 {
702 	struct sock *sk, *first = NULL;
703 	const struct udphdr *uh = udp_hdr(skb);
704 	unsigned short hnum = ntohs(uh->dest);
705 	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
706 	unsigned int offset = offsetof(typeof(*sk), sk_node);
707 	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
708 	int dif = inet6_iif(skb);
709 	struct hlist_node *node;
710 	struct sk_buff *nskb;
711 
712 	if (use_hash2) {
713 		hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) &
714 			    udptable->mask;
715 		hash2 = udp6_portaddr_hash(net, daddr, hnum) & udptable->mask;
716 start_lookup:
717 		hslot = &udptable->hash2[hash2];
718 		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
719 	}
720 
721 	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
722 		if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
723 					    uh->source, saddr, dif, hnum))
724 			continue;
725 		/* If zero checksum and no_check is not on for
726 		 * the socket then skip it.
727 		 */
728 		if (!uh->check && !udp_sk(sk)->no_check6_rx)
729 			continue;
730 		if (!first) {
731 			first = sk;
732 			continue;
733 		}
734 		nskb = skb_clone(skb, GFP_ATOMIC);
735 		if (unlikely(!nskb)) {
736 			atomic_inc(&sk->sk_drops);
737 			__UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
738 					 IS_UDPLITE(sk));
739 			__UDP6_INC_STATS(net, UDP_MIB_INERRORS,
740 					 IS_UDPLITE(sk));
741 			continue;
742 		}
743 
744 		if (udpv6_queue_rcv_skb(sk, nskb) > 0)
745 			consume_skb(nskb);
746 	}
747 
748 	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
749 	if (use_hash2 && hash2 != hash2_any) {
750 		hash2 = hash2_any;
751 		goto start_lookup;
752 	}
753 
754 	if (first) {
755 		if (udpv6_queue_rcv_skb(first, skb) > 0)
756 			consume_skb(skb);
757 	} else {
758 		kfree_skb(skb);
759 		__UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
760 				 proto == IPPROTO_UDPLITE);
761 	}
762 	return 0;
763 }
764 
765 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
766 		   int proto)
767 {
768 	const struct in6_addr *saddr, *daddr;
769 	struct net *net = dev_net(skb->dev);
770 	struct udphdr *uh;
771 	struct sock *sk;
772 	u32 ulen = 0;
773 
774 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
775 		goto discard;
776 
777 	saddr = &ipv6_hdr(skb)->saddr;
778 	daddr = &ipv6_hdr(skb)->daddr;
779 	uh = udp_hdr(skb);
780 
781 	ulen = ntohs(uh->len);
782 	if (ulen > skb->len)
783 		goto short_packet;
784 
785 	if (proto == IPPROTO_UDP) {
786 		/* UDP validates ulen. */
787 
788 		/* Check for jumbo payload */
789 		if (ulen == 0)
790 			ulen = skb->len;
791 
792 		if (ulen < sizeof(*uh))
793 			goto short_packet;
794 
795 		if (ulen < skb->len) {
796 			if (pskb_trim_rcsum(skb, ulen))
797 				goto short_packet;
798 			saddr = &ipv6_hdr(skb)->saddr;
799 			daddr = &ipv6_hdr(skb)->daddr;
800 			uh = udp_hdr(skb);
801 		}
802 	}
803 
804 	if (udp6_csum_init(skb, uh, proto))
805 		goto csum_error;
806 
807 	/*
808 	 *	Multicast receive code
809 	 */
810 	if (ipv6_addr_is_multicast(daddr))
811 		return __udp6_lib_mcast_deliver(net, skb,
812 				saddr, daddr, udptable, proto);
813 
814 	/* Unicast */
815 
816 	/*
817 	 * check socket cache ... must talk to Alan about his plans
818 	 * for sock caches... i'll skip this for now.
819 	 */
820 	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
821 	if (sk) {
822 		int ret;
823 
824 		if (!uh->check && !udp_sk(sk)->no_check6_rx) {
825 			udp6_csum_zero_error(skb);
826 			goto csum_error;
827 		}
828 
829 		if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
830 			skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
831 						 ip6_compute_pseudo);
832 
833 		ret = udpv6_queue_rcv_skb(sk, skb);
834 
835 		/* a return value > 0 means to resubmit the input */
836 		if (ret > 0)
837 			return ret;
838 
839 		return 0;
840 	}
841 
842 	if (!uh->check) {
843 		udp6_csum_zero_error(skb);
844 		goto csum_error;
845 	}
846 
847 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
848 		goto discard;
849 
850 	if (udp_lib_checksum_complete(skb))
851 		goto csum_error;
852 
853 	__UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
854 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
855 
856 	kfree_skb(skb);
857 	return 0;
858 
859 short_packet:
860 	net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
861 			    proto == IPPROTO_UDPLITE ? "-Lite" : "",
862 			    saddr, ntohs(uh->source),
863 			    ulen, skb->len,
864 			    daddr, ntohs(uh->dest));
865 	goto discard;
866 csum_error:
867 	__UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
868 discard:
869 	__UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
870 	kfree_skb(skb);
871 	return 0;
872 }
873 
874 
875 static struct sock *__udp6_lib_demux_lookup(struct net *net,
876 			__be16 loc_port, const struct in6_addr *loc_addr,
877 			__be16 rmt_port, const struct in6_addr *rmt_addr,
878 			int dif)
879 {
880 	unsigned short hnum = ntohs(loc_port);
881 	unsigned int hash2 = udp6_portaddr_hash(net, loc_addr, hnum);
882 	unsigned int slot2 = hash2 & udp_table.mask;
883 	struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
884 	const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
885 	struct sock *sk;
886 
887 	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
888 		if (sk->sk_state == TCP_ESTABLISHED &&
889 		    INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif))
890 			return sk;
891 		/* Only check first socket in chain */
892 		break;
893 	}
894 	return NULL;
895 }
896 
897 static void udp_v6_early_demux(struct sk_buff *skb)
898 {
899 	struct net *net = dev_net(skb->dev);
900 	const struct udphdr *uh;
901 	struct sock *sk;
902 	struct dst_entry *dst;
903 	int dif = skb->dev->ifindex;
904 
905 	if (!pskb_may_pull(skb, skb_transport_offset(skb) +
906 	    sizeof(struct udphdr)))
907 		return;
908 
909 	uh = udp_hdr(skb);
910 
911 	if (skb->pkt_type == PACKET_HOST)
912 		sk = __udp6_lib_demux_lookup(net, uh->dest,
913 					     &ipv6_hdr(skb)->daddr,
914 					     uh->source, &ipv6_hdr(skb)->saddr,
915 					     dif);
916 	else
917 		return;
918 
919 	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
920 		return;
921 
922 	skb->sk = sk;
923 	skb->destructor = sock_efree;
924 	dst = READ_ONCE(sk->sk_rx_dst);
925 
926 	if (dst)
927 		dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
928 	if (dst) {
929 		/* set noref for now.
930 		 * any place which wants to hold dst has to call
931 		 * dst_hold_safe()
932 		 */
933 		skb_dst_set_noref(skb, dst);
934 	}
935 }
936 
937 static __inline__ int udpv6_rcv(struct sk_buff *skb)
938 {
939 	return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
940 }
941 
942 /*
943  * Throw away all pending data and cancel the corking. Socket is locked.
944  */
945 static void udp_v6_flush_pending_frames(struct sock *sk)
946 {
947 	struct udp_sock *up = udp_sk(sk);
948 
949 	if (up->pending == AF_INET)
950 		udp_flush_pending_frames(sk);
951 	else if (up->pending) {
952 		up->len = 0;
953 		up->pending = 0;
954 		ip6_flush_pending_frames(sk);
955 	}
956 }
957 
958 /**
959  *	udp6_hwcsum_outgoing  -  handle outgoing HW checksumming
960  *	@sk:	socket we are sending on
961  *	@skb:	sk_buff containing the filled-in UDP header
962  *		(checksum field must be zeroed out)
963  */
964 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
965 				 const struct in6_addr *saddr,
966 				 const struct in6_addr *daddr, int len)
967 {
968 	unsigned int offset;
969 	struct udphdr *uh = udp_hdr(skb);
970 	struct sk_buff *frags = skb_shinfo(skb)->frag_list;
971 	__wsum csum = 0;
972 
973 	if (!frags) {
974 		/* Only one fragment on the socket.  */
975 		skb->csum_start = skb_transport_header(skb) - skb->head;
976 		skb->csum_offset = offsetof(struct udphdr, check);
977 		uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
978 	} else {
979 		/*
980 		 * HW-checksum won't work as there are two or more
981 		 * fragments on the socket so that all csums of sk_buffs
982 		 * should be together
983 		 */
984 		offset = skb_transport_offset(skb);
985 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
986 
987 		skb->ip_summed = CHECKSUM_NONE;
988 
989 		do {
990 			csum = csum_add(csum, frags->csum);
991 		} while ((frags = frags->next));
992 
993 		uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
994 					    csum);
995 		if (uh->check == 0)
996 			uh->check = CSUM_MANGLED_0;
997 	}
998 }
999 
1000 /*
1001  *	Sending
1002  */
1003 
1004 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6)
1005 {
1006 	struct sock *sk = skb->sk;
1007 	struct udphdr *uh;
1008 	int err = 0;
1009 	int is_udplite = IS_UDPLITE(sk);
1010 	__wsum csum = 0;
1011 	int offset = skb_transport_offset(skb);
1012 	int len = skb->len - offset;
1013 
1014 	/*
1015 	 * Create a UDP header
1016 	 */
1017 	uh = udp_hdr(skb);
1018 	uh->source = fl6->fl6_sport;
1019 	uh->dest = fl6->fl6_dport;
1020 	uh->len = htons(len);
1021 	uh->check = 0;
1022 
1023 	if (is_udplite)
1024 		csum = udplite_csum(skb);
1025 	else if (udp_sk(sk)->no_check6_tx) {   /* UDP csum disabled */
1026 		skb->ip_summed = CHECKSUM_NONE;
1027 		goto send;
1028 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1029 		udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1030 		goto send;
1031 	} else
1032 		csum = udp_csum(skb);
1033 
1034 	/* add protocol-dependent pseudo-header */
1035 	uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1036 				    len, fl6->flowi6_proto, csum);
1037 	if (uh->check == 0)
1038 		uh->check = CSUM_MANGLED_0;
1039 
1040 send:
1041 	err = ip6_send_skb(skb);
1042 	if (err) {
1043 		if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1044 			UDP6_INC_STATS(sock_net(sk),
1045 				       UDP_MIB_SNDBUFERRORS, is_udplite);
1046 			err = 0;
1047 		}
1048 	} else {
1049 		UDP6_INC_STATS(sock_net(sk),
1050 			       UDP_MIB_OUTDATAGRAMS, is_udplite);
1051 	}
1052 	return err;
1053 }
1054 
1055 static int udp_v6_push_pending_frames(struct sock *sk)
1056 {
1057 	struct sk_buff *skb;
1058 	struct udp_sock  *up = udp_sk(sk);
1059 	struct flowi6 fl6;
1060 	int err = 0;
1061 
1062 	if (up->pending == AF_INET)
1063 		return udp_push_pending_frames(sk);
1064 
1065 	/* ip6_finish_skb will release the cork, so make a copy of
1066 	 * fl6 here.
1067 	 */
1068 	fl6 = inet_sk(sk)->cork.fl.u.ip6;
1069 
1070 	skb = ip6_finish_skb(sk);
1071 	if (!skb)
1072 		goto out;
1073 
1074 	err = udp_v6_send_skb(skb, &fl6);
1075 
1076 out:
1077 	up->len = 0;
1078 	up->pending = 0;
1079 	return err;
1080 }
1081 
1082 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1083 {
1084 	struct ipv6_txoptions opt_space;
1085 	struct udp_sock *up = udp_sk(sk);
1086 	struct inet_sock *inet = inet_sk(sk);
1087 	struct ipv6_pinfo *np = inet6_sk(sk);
1088 	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1089 	struct in6_addr *daddr, *final_p, final;
1090 	struct ipv6_txoptions *opt = NULL;
1091 	struct ipv6_txoptions *opt_to_free = NULL;
1092 	struct ip6_flowlabel *flowlabel = NULL;
1093 	struct flowi6 fl6;
1094 	struct dst_entry *dst;
1095 	struct ipcm6_cookie ipc6;
1096 	int addr_len = msg->msg_namelen;
1097 	int ulen = len;
1098 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1099 	int err;
1100 	int connected = 0;
1101 	int is_udplite = IS_UDPLITE(sk);
1102 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1103 	struct sockcm_cookie sockc;
1104 
1105 	ipc6.hlimit = -1;
1106 	ipc6.tclass = -1;
1107 	ipc6.dontfrag = -1;
1108 	sockc.tsflags = sk->sk_tsflags;
1109 
1110 	/* destination address check */
1111 	if (sin6) {
1112 		if (addr_len < offsetof(struct sockaddr, sa_data))
1113 			return -EINVAL;
1114 
1115 		switch (sin6->sin6_family) {
1116 		case AF_INET6:
1117 			if (addr_len < SIN6_LEN_RFC2133)
1118 				return -EINVAL;
1119 			daddr = &sin6->sin6_addr;
1120 			if (ipv6_addr_any(daddr) &&
1121 			    ipv6_addr_v4mapped(&np->saddr))
1122 				ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1123 						       daddr);
1124 			break;
1125 		case AF_INET:
1126 			goto do_udp_sendmsg;
1127 		case AF_UNSPEC:
1128 			msg->msg_name = sin6 = NULL;
1129 			msg->msg_namelen = addr_len = 0;
1130 			daddr = NULL;
1131 			break;
1132 		default:
1133 			return -EINVAL;
1134 		}
1135 	} else if (!up->pending) {
1136 		if (sk->sk_state != TCP_ESTABLISHED)
1137 			return -EDESTADDRREQ;
1138 		daddr = &sk->sk_v6_daddr;
1139 	} else
1140 		daddr = NULL;
1141 
1142 	if (daddr) {
1143 		if (ipv6_addr_v4mapped(daddr)) {
1144 			struct sockaddr_in sin;
1145 			sin.sin_family = AF_INET;
1146 			sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1147 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
1148 			msg->msg_name = &sin;
1149 			msg->msg_namelen = sizeof(sin);
1150 do_udp_sendmsg:
1151 			if (__ipv6_only_sock(sk))
1152 				return -ENETUNREACH;
1153 			return udp_sendmsg(sk, msg, len);
1154 		}
1155 	}
1156 
1157 	if (up->pending == AF_INET)
1158 		return udp_sendmsg(sk, msg, len);
1159 
1160 	/* Rough check on arithmetic overflow,
1161 	   better check is made in ip6_append_data().
1162 	   */
1163 	if (len > INT_MAX - sizeof(struct udphdr))
1164 		return -EMSGSIZE;
1165 
1166 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
1167 	if (up->pending) {
1168 		/*
1169 		 * There are pending frames.
1170 		 * The socket lock must be held while it's corked.
1171 		 */
1172 		lock_sock(sk);
1173 		if (likely(up->pending)) {
1174 			if (unlikely(up->pending != AF_INET6)) {
1175 				release_sock(sk);
1176 				return -EAFNOSUPPORT;
1177 			}
1178 			dst = NULL;
1179 			goto do_append_data;
1180 		}
1181 		release_sock(sk);
1182 	}
1183 	ulen += sizeof(struct udphdr);
1184 
1185 	memset(&fl6, 0, sizeof(fl6));
1186 
1187 	if (sin6) {
1188 		if (sin6->sin6_port == 0)
1189 			return -EINVAL;
1190 
1191 		fl6.fl6_dport = sin6->sin6_port;
1192 		daddr = &sin6->sin6_addr;
1193 
1194 		if (np->sndflow) {
1195 			fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1196 			if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1197 				flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1198 				if (!flowlabel)
1199 					return -EINVAL;
1200 			}
1201 		}
1202 
1203 		/*
1204 		 * Otherwise it will be difficult to maintain
1205 		 * sk->sk_dst_cache.
1206 		 */
1207 		if (sk->sk_state == TCP_ESTABLISHED &&
1208 		    ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1209 			daddr = &sk->sk_v6_daddr;
1210 
1211 		if (addr_len >= sizeof(struct sockaddr_in6) &&
1212 		    sin6->sin6_scope_id &&
1213 		    __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1214 			fl6.flowi6_oif = sin6->sin6_scope_id;
1215 	} else {
1216 		if (sk->sk_state != TCP_ESTABLISHED)
1217 			return -EDESTADDRREQ;
1218 
1219 		fl6.fl6_dport = inet->inet_dport;
1220 		daddr = &sk->sk_v6_daddr;
1221 		fl6.flowlabel = np->flow_label;
1222 		connected = 1;
1223 	}
1224 
1225 	if (!fl6.flowi6_oif)
1226 		fl6.flowi6_oif = sk->sk_bound_dev_if;
1227 
1228 	if (!fl6.flowi6_oif)
1229 		fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1230 
1231 	fl6.flowi6_mark = sk->sk_mark;
1232 	fl6.flowi6_uid = sk->sk_uid;
1233 
1234 	if (msg->msg_controllen) {
1235 		opt = &opt_space;
1236 		memset(opt, 0, sizeof(struct ipv6_txoptions));
1237 		opt->tot_len = sizeof(*opt);
1238 		ipc6.opt = opt;
1239 
1240 		err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6, &sockc);
1241 		if (err < 0) {
1242 			fl6_sock_release(flowlabel);
1243 			return err;
1244 		}
1245 		if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1246 			flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1247 			if (!flowlabel)
1248 				return -EINVAL;
1249 		}
1250 		if (!(opt->opt_nflen|opt->opt_flen))
1251 			opt = NULL;
1252 		connected = 0;
1253 	}
1254 	if (!opt) {
1255 		opt = txopt_get(np);
1256 		opt_to_free = opt;
1257 	}
1258 	if (flowlabel)
1259 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
1260 	opt = ipv6_fixup_options(&opt_space, opt);
1261 	ipc6.opt = opt;
1262 
1263 	fl6.flowi6_proto = sk->sk_protocol;
1264 	if (!ipv6_addr_any(daddr))
1265 		fl6.daddr = *daddr;
1266 	else
1267 		fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1268 	if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1269 		fl6.saddr = np->saddr;
1270 	fl6.fl6_sport = inet->inet_sport;
1271 
1272 	final_p = fl6_update_dst(&fl6, opt, &final);
1273 	if (final_p)
1274 		connected = 0;
1275 
1276 	if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1277 		fl6.flowi6_oif = np->mcast_oif;
1278 		connected = 0;
1279 	} else if (!fl6.flowi6_oif)
1280 		fl6.flowi6_oif = np->ucast_oif;
1281 
1282 	security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1283 
1284 	if (ipc6.tclass < 0)
1285 		ipc6.tclass = np->tclass;
1286 
1287 	fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1288 
1289 	dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p);
1290 	if (IS_ERR(dst)) {
1291 		err = PTR_ERR(dst);
1292 		dst = NULL;
1293 		goto out;
1294 	}
1295 
1296 	if (ipc6.hlimit < 0)
1297 		ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1298 
1299 	if (msg->msg_flags&MSG_CONFIRM)
1300 		goto do_confirm;
1301 back_from_confirm:
1302 
1303 	/* Lockless fast path for the non-corking case */
1304 	if (!corkreq) {
1305 		struct sk_buff *skb;
1306 
1307 		skb = ip6_make_skb(sk, getfrag, msg, ulen,
1308 				   sizeof(struct udphdr), &ipc6,
1309 				   &fl6, (struct rt6_info *)dst,
1310 				   msg->msg_flags, &sockc);
1311 		err = PTR_ERR(skb);
1312 		if (!IS_ERR_OR_NULL(skb))
1313 			err = udp_v6_send_skb(skb, &fl6);
1314 		goto release_dst;
1315 	}
1316 
1317 	lock_sock(sk);
1318 	if (unlikely(up->pending)) {
1319 		/* The socket is already corked while preparing it. */
1320 		/* ... which is an evident application bug. --ANK */
1321 		release_sock(sk);
1322 
1323 		net_dbg_ratelimited("udp cork app bug 2\n");
1324 		err = -EINVAL;
1325 		goto out;
1326 	}
1327 
1328 	up->pending = AF_INET6;
1329 
1330 do_append_data:
1331 	if (ipc6.dontfrag < 0)
1332 		ipc6.dontfrag = np->dontfrag;
1333 	up->len += ulen;
1334 	err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1335 			      &ipc6, &fl6, (struct rt6_info *)dst,
1336 			      corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, &sockc);
1337 	if (err)
1338 		udp_v6_flush_pending_frames(sk);
1339 	else if (!corkreq)
1340 		err = udp_v6_push_pending_frames(sk);
1341 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1342 		up->pending = 0;
1343 
1344 	if (err > 0)
1345 		err = np->recverr ? net_xmit_errno(err) : 0;
1346 	release_sock(sk);
1347 
1348 release_dst:
1349 	if (dst) {
1350 		if (connected) {
1351 			ip6_dst_store(sk, dst,
1352 				      ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
1353 				      &sk->sk_v6_daddr : NULL,
1354 #ifdef CONFIG_IPV6_SUBTREES
1355 				      ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1356 				      &np->saddr :
1357 #endif
1358 				      NULL);
1359 		} else {
1360 			dst_release(dst);
1361 		}
1362 		dst = NULL;
1363 	}
1364 
1365 out:
1366 	dst_release(dst);
1367 	fl6_sock_release(flowlabel);
1368 	txopt_put(opt_to_free);
1369 	if (!err)
1370 		return len;
1371 	/*
1372 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1373 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1374 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1375 	 * things).  We could add another new stat but at least for now that
1376 	 * seems like overkill.
1377 	 */
1378 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1379 		UDP6_INC_STATS(sock_net(sk),
1380 			       UDP_MIB_SNDBUFERRORS, is_udplite);
1381 	}
1382 	return err;
1383 
1384 do_confirm:
1385 	if (msg->msg_flags & MSG_PROBE)
1386 		dst_confirm_neigh(dst, &fl6.daddr);
1387 	if (!(msg->msg_flags&MSG_PROBE) || len)
1388 		goto back_from_confirm;
1389 	err = 0;
1390 	goto out;
1391 }
1392 
1393 void udpv6_destroy_sock(struct sock *sk)
1394 {
1395 	struct udp_sock *up = udp_sk(sk);
1396 	lock_sock(sk);
1397 	udp_v6_flush_pending_frames(sk);
1398 	release_sock(sk);
1399 
1400 	if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
1401 		void (*encap_destroy)(struct sock *sk);
1402 		encap_destroy = ACCESS_ONCE(up->encap_destroy);
1403 		if (encap_destroy)
1404 			encap_destroy(sk);
1405 	}
1406 
1407 	inet6_destroy_sock(sk);
1408 }
1409 
1410 /*
1411  *	Socket option code for UDP
1412  */
1413 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1414 		     char __user *optval, unsigned int optlen)
1415 {
1416 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1417 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1418 					  udp_v6_push_pending_frames);
1419 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
1420 }
1421 
1422 #ifdef CONFIG_COMPAT
1423 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1424 			    char __user *optval, unsigned int optlen)
1425 {
1426 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1427 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1428 					  udp_v6_push_pending_frames);
1429 	return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1430 }
1431 #endif
1432 
1433 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1434 		     char __user *optval, int __user *optlen)
1435 {
1436 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1437 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1438 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
1439 }
1440 
1441 #ifdef CONFIG_COMPAT
1442 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1443 			    char __user *optval, int __user *optlen)
1444 {
1445 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1446 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1447 	return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1448 }
1449 #endif
1450 
1451 static struct inet6_protocol udpv6_protocol = {
1452 	.early_demux	=	udp_v6_early_demux,
1453 	.early_demux_handler =  udp_v6_early_demux,
1454 	.handler	=	udpv6_rcv,
1455 	.err_handler	=	udpv6_err,
1456 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1457 };
1458 
1459 /* ------------------------------------------------------------------------ */
1460 #ifdef CONFIG_PROC_FS
1461 int udp6_seq_show(struct seq_file *seq, void *v)
1462 {
1463 	if (v == SEQ_START_TOKEN) {
1464 		seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1465 	} else {
1466 		int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1467 		struct inet_sock *inet = inet_sk(v);
1468 		__u16 srcp = ntohs(inet->inet_sport);
1469 		__u16 destp = ntohs(inet->inet_dport);
1470 		ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket);
1471 	}
1472 	return 0;
1473 }
1474 
1475 static const struct file_operations udp6_afinfo_seq_fops = {
1476 	.owner    = THIS_MODULE,
1477 	.open     = udp_seq_open,
1478 	.read     = seq_read,
1479 	.llseek   = seq_lseek,
1480 	.release  = seq_release_net
1481 };
1482 
1483 static struct udp_seq_afinfo udp6_seq_afinfo = {
1484 	.name		= "udp6",
1485 	.family		= AF_INET6,
1486 	.udp_table	= &udp_table,
1487 	.seq_fops	= &udp6_afinfo_seq_fops,
1488 	.seq_ops	= {
1489 		.show		= udp6_seq_show,
1490 	},
1491 };
1492 
1493 int __net_init udp6_proc_init(struct net *net)
1494 {
1495 	return udp_proc_register(net, &udp6_seq_afinfo);
1496 }
1497 
1498 void udp6_proc_exit(struct net *net)
1499 {
1500 	udp_proc_unregister(net, &udp6_seq_afinfo);
1501 }
1502 #endif /* CONFIG_PROC_FS */
1503 
1504 /* ------------------------------------------------------------------------ */
1505 
1506 struct proto udpv6_prot = {
1507 	.name		   = "UDPv6",
1508 	.owner		   = THIS_MODULE,
1509 	.close		   = udp_lib_close,
1510 	.connect	   = ip6_datagram_connect,
1511 	.disconnect	   = udp_disconnect,
1512 	.ioctl		   = udp_ioctl,
1513 	.init		   = udp_init_sock,
1514 	.destroy	   = udpv6_destroy_sock,
1515 	.setsockopt	   = udpv6_setsockopt,
1516 	.getsockopt	   = udpv6_getsockopt,
1517 	.sendmsg	   = udpv6_sendmsg,
1518 	.recvmsg	   = udpv6_recvmsg,
1519 	.release_cb	   = ip6_datagram_release_cb,
1520 	.hash		   = udp_lib_hash,
1521 	.unhash		   = udp_lib_unhash,
1522 	.rehash		   = udp_v6_rehash,
1523 	.get_port	   = udp_v6_get_port,
1524 	.memory_allocated  = &udp_memory_allocated,
1525 	.sysctl_mem	   = sysctl_udp_mem,
1526 	.sysctl_wmem	   = &sysctl_udp_wmem_min,
1527 	.sysctl_rmem	   = &sysctl_udp_rmem_min,
1528 	.obj_size	   = sizeof(struct udp6_sock),
1529 	.h.udp_table	   = &udp_table,
1530 #ifdef CONFIG_COMPAT
1531 	.compat_setsockopt = compat_udpv6_setsockopt,
1532 	.compat_getsockopt = compat_udpv6_getsockopt,
1533 #endif
1534 	.diag_destroy      = udp_abort,
1535 };
1536 
1537 static struct inet_protosw udpv6_protosw = {
1538 	.type =      SOCK_DGRAM,
1539 	.protocol =  IPPROTO_UDP,
1540 	.prot =      &udpv6_prot,
1541 	.ops =       &inet6_dgram_ops,
1542 	.flags =     INET_PROTOSW_PERMANENT,
1543 };
1544 
1545 int __init udpv6_init(void)
1546 {
1547 	int ret;
1548 
1549 	ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1550 	if (ret)
1551 		goto out;
1552 
1553 	ret = inet6_register_protosw(&udpv6_protosw);
1554 	if (ret)
1555 		goto out_udpv6_protocol;
1556 out:
1557 	return ret;
1558 
1559 out_udpv6_protocol:
1560 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1561 	goto out;
1562 }
1563 
1564 void udpv6_exit(void)
1565 {
1566 	inet6_unregister_protosw(&udpv6_protosw);
1567 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1568 }
1569