xref: /linux/net/ipv6/udp.c (revision 6dfafbd0299a60bfb5d5e277fdf100037c7ded07)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	UDP over IPv6
4  *	Linux INET6 implementation
5  *
6  *	Authors:
7  *	Pedro Roque		<roque@di.fc.ul.pt>
8  *
9  *	Based on linux/ipv4/udp.c
10  *
11  *	Fixes:
12  *	Hideaki YOSHIFUJI	:	sin6_scope_id support
13  *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
14  *	Alexey Kuznetsov		allow both IPv4 and IPv6 sockets to bind
15  *					a single port at the same time.
16  *      Kazunori MIYAZAWA @USAGI:       change process style to use ip6_append_data
17  *      YOSHIFUJI Hideaki @USAGI:	convert /proc/net/udp6 to seq_file.
18  */
19 
20 #include <linux/bpf-cgroup.h>
21 #include <linux/errno.h>
22 #include <linux/types.h>
23 #include <linux/socket.h>
24 #include <linux/sockios.h>
25 #include <linux/net.h>
26 #include <linux/in6.h>
27 #include <linux/netdevice.h>
28 #include <linux/if_arp.h>
29 #include <linux/ipv6.h>
30 #include <linux/icmpv6.h>
31 #include <linux/init.h>
32 #include <linux/module.h>
33 #include <linux/skbuff.h>
34 #include <linux/slab.h>
35 #include <linux/uaccess.h>
36 #include <linux/indirect_call_wrapper.h>
37 #include <trace/events/udp.h>
38 
39 #include <net/addrconf.h>
40 #include <net/ndisc.h>
41 #include <net/protocol.h>
42 #include <net/transp_v6.h>
43 #include <net/ip6_route.h>
44 #include <net/raw.h>
45 #include <net/seg6.h>
46 #include <net/tcp_states.h>
47 #include <net/ip6_checksum.h>
48 #include <net/ip6_tunnel.h>
49 #include <net/udp_tunnel.h>
50 #include <net/xfrm.h>
51 #include <net/inet_hashtables.h>
52 #include <net/inet6_hashtables.h>
53 #include <net/busy_poll.h>
54 #include <net/sock_reuseport.h>
55 #include <net/gro.h>
56 
57 #include <linux/proc_fs.h>
58 #include <linux/seq_file.h>
59 #include <trace/events/skb.h>
60 #include "udp_impl.h"
61 
62 static void udpv6_destruct_sock(struct sock *sk)
63 {
64 	udp_destruct_common(sk);
65 	inet6_sock_destruct(sk);
66 }
67 
68 int udpv6_init_sock(struct sock *sk)
69 {
70 	int res = udp_lib_init_sock(sk);
71 
72 	sk->sk_destruct = udpv6_destruct_sock;
73 	set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
74 	return res;
75 }
76 
77 INDIRECT_CALLABLE_SCOPE
78 u32 udp6_ehashfn(const struct net *net,
79 		 const struct in6_addr *laddr,
80 		 const u16 lport,
81 		 const struct in6_addr *faddr,
82 		 const __be16 fport)
83 {
84 	u32 lhash, fhash;
85 
86 	net_get_random_once(&udp6_ehash_secret,
87 			    sizeof(udp6_ehash_secret));
88 	net_get_random_once(&udp_ipv6_hash_secret,
89 			    sizeof(udp_ipv6_hash_secret));
90 
91 	lhash = (__force u32)laddr->s6_addr32[3];
92 	fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
93 
94 	return __inet6_ehashfn(lhash, lport, fhash, fport,
95 			       udp6_ehash_secret + net_hash_mix(net));
96 }
97 
98 int udp_v6_get_port(struct sock *sk, unsigned short snum)
99 {
100 	unsigned int hash2_nulladdr =
101 		ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
102 	unsigned int hash2_partial =
103 		ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
104 
105 	/* precompute partial secondary hash */
106 	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
107 	return udp_lib_get_port(sk, snum, hash2_nulladdr);
108 }
109 
110 void udp_v6_rehash(struct sock *sk)
111 {
112 	u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
113 					  &sk->sk_v6_rcv_saddr,
114 					  inet_sk(sk)->inet_num);
115 	u16 new_hash4;
116 
117 	if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) {
118 		new_hash4 = udp_ehashfn(sock_net(sk),
119 					sk->sk_rcv_saddr, sk->sk_num,
120 					sk->sk_daddr, sk->sk_dport);
121 	} else {
122 		new_hash4 = udp6_ehashfn(sock_net(sk),
123 					 &sk->sk_v6_rcv_saddr, sk->sk_num,
124 					 &sk->sk_v6_daddr, sk->sk_dport);
125 	}
126 
127 	udp_lib_rehash(sk, new_hash, new_hash4);
128 }
129 
130 static int compute_score(struct sock *sk, const struct net *net,
131 			 const struct in6_addr *saddr, __be16 sport,
132 			 const struct in6_addr *daddr, unsigned short hnum,
133 			 int dif, int sdif)
134 {
135 	int bound_dev_if, score;
136 	struct inet_sock *inet;
137 	bool dev_match;
138 
139 	if (!net_eq(sock_net(sk), net) ||
140 	    udp_sk(sk)->udp_port_hash != hnum ||
141 	    sk->sk_family != PF_INET6)
142 		return -1;
143 
144 	if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
145 		return -1;
146 
147 	score = 0;
148 	inet = inet_sk(sk);
149 
150 	if (inet->inet_dport) {
151 		if (inet->inet_dport != sport)
152 			return -1;
153 		score++;
154 	}
155 
156 	if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
157 		if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
158 			return -1;
159 		score++;
160 	}
161 
162 	bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
163 	dev_match = udp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif);
164 	if (!dev_match)
165 		return -1;
166 	if (bound_dev_if)
167 		score++;
168 
169 	if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
170 		score++;
171 
172 	return score;
173 }
174 
175 /**
176  * udp6_lib_lookup1() - Simplified lookup using primary hash (destination port)
177  * @net:	Network namespace
178  * @saddr:	Source address, network order
179  * @sport:	Source port, network order
180  * @daddr:	Destination address, network order
181  * @hnum:	Destination port, host order
182  * @dif:	Destination interface index
183  * @sdif:	Destination bridge port index, if relevant
184  * @udptable:	Set of UDP hash tables
185  *
186  * Simplified lookup to be used as fallback if no sockets are found due to a
187  * potential race between (receive) address change, and lookup happening before
188  * the rehash operation. This function ignores SO_REUSEPORT groups while scoring
189  * result sockets, because if we have one, we don't need the fallback at all.
190  *
191  * Called under rcu_read_lock().
192  *
193  * Return: socket with highest matching score if any, NULL if none
194  */
195 static struct sock *udp6_lib_lookup1(const struct net *net,
196 				     const struct in6_addr *saddr, __be16 sport,
197 				     const struct in6_addr *daddr,
198 				     unsigned int hnum, int dif, int sdif,
199 				     const struct udp_table *udptable)
200 {
201 	unsigned int slot = udp_hashfn(net, hnum, udptable->mask);
202 	struct udp_hslot *hslot = &udptable->hash[slot];
203 	struct sock *sk, *result = NULL;
204 	int score, badness = 0;
205 
206 	sk_for_each_rcu(sk, &hslot->head) {
207 		score = compute_score(sk, net,
208 				      saddr, sport, daddr, hnum, dif, sdif);
209 		if (score > badness) {
210 			result = sk;
211 			badness = score;
212 		}
213 	}
214 
215 	return result;
216 }
217 
218 /* called with rcu_read_lock() */
219 static struct sock *udp6_lib_lookup2(const struct net *net,
220 		const struct in6_addr *saddr, __be16 sport,
221 		const struct in6_addr *daddr, unsigned int hnum,
222 		int dif, int sdif, struct udp_hslot *hslot2,
223 		struct sk_buff *skb)
224 {
225 	struct sock *sk, *result;
226 	int score, badness;
227 	bool need_rescore;
228 
229 	result = NULL;
230 	badness = -1;
231 	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
232 		need_rescore = false;
233 rescore:
234 		score = compute_score(need_rescore ? result : sk, net, saddr,
235 				      sport, daddr, hnum, dif, sdif);
236 		if (score > badness) {
237 			badness = score;
238 
239 			if (need_rescore)
240 				continue;
241 
242 			if (sk->sk_state == TCP_ESTABLISHED) {
243 				result = sk;
244 				continue;
245 			}
246 
247 			result = inet6_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
248 							saddr, sport, daddr, hnum, udp6_ehashfn);
249 			if (!result) {
250 				result = sk;
251 				continue;
252 			}
253 
254 			/* Fall back to scoring if group has connections */
255 			if (!reuseport_has_conns(sk))
256 				return result;
257 
258 			/* Reuseport logic returned an error, keep original score. */
259 			if (IS_ERR(result))
260 				continue;
261 
262 			/* compute_score is too long of a function to be
263 			 * inlined, and calling it again here yields
264 			 * measurable overhead for some
265 			 * workloads. Work around it by jumping
266 			 * backwards to rescore 'result'.
267 			 */
268 			need_rescore = true;
269 			goto rescore;
270 		}
271 	}
272 	return result;
273 }
274 
275 #if IS_ENABLED(CONFIG_BASE_SMALL)
276 static struct sock *udp6_lib_lookup4(const struct net *net,
277 				     const struct in6_addr *saddr, __be16 sport,
278 				     const struct in6_addr *daddr,
279 				     unsigned int hnum, int dif, int sdif,
280 				     struct udp_table *udptable)
281 {
282 	return NULL;
283 }
284 
285 static void udp6_hash4(struct sock *sk)
286 {
287 }
288 #else /* !CONFIG_BASE_SMALL */
289 static struct sock *udp6_lib_lookup4(const struct net *net,
290 				     const struct in6_addr *saddr, __be16 sport,
291 				     const struct in6_addr *daddr,
292 				     unsigned int hnum, int dif, int sdif,
293 				     struct udp_table *udptable)
294 {
295 	const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
296 	const struct hlist_nulls_node *node;
297 	struct udp_hslot *hslot4;
298 	unsigned int hash4, slot;
299 	struct udp_sock *up;
300 	struct sock *sk;
301 
302 	hash4 = udp6_ehashfn(net, daddr, hnum, saddr, sport);
303 	slot = hash4 & udptable->mask;
304 	hslot4 = &udptable->hash4[slot];
305 
306 begin:
307 	udp_lrpa_for_each_entry_rcu(up, node, &hslot4->nulls_head) {
308 		sk = (struct sock *)up;
309 		if (inet6_match(net, sk, saddr, daddr, ports, dif, sdif))
310 			return sk;
311 	}
312 
313 	/* if the nulls value we got at the end of this lookup is not the
314 	 * expected one, we must restart lookup. We probably met an item that
315 	 * was moved to another chain due to rehash.
316 	 */
317 	if (get_nulls_value(node) != slot)
318 		goto begin;
319 
320 	return NULL;
321 }
322 
323 static void udp6_hash4(struct sock *sk)
324 {
325 	struct net *net = sock_net(sk);
326 	unsigned int hash;
327 
328 	if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) {
329 		udp4_hash4(sk);
330 		return;
331 	}
332 
333 	if (sk_unhashed(sk) || ipv6_addr_any(&sk->sk_v6_rcv_saddr))
334 		return;
335 
336 	hash = udp6_ehashfn(net, &sk->sk_v6_rcv_saddr, sk->sk_num,
337 			    &sk->sk_v6_daddr, sk->sk_dport);
338 
339 	udp_lib_hash4(sk, hash);
340 }
341 #endif /* CONFIG_BASE_SMALL */
342 
343 /* rcu_read_lock() must be held */
344 struct sock *__udp6_lib_lookup(const struct net *net,
345 			       const struct in6_addr *saddr, __be16 sport,
346 			       const struct in6_addr *daddr, __be16 dport,
347 			       int dif, int sdif, struct udp_table *udptable,
348 			       struct sk_buff *skb)
349 {
350 	unsigned short hnum = ntohs(dport);
351 	struct udp_hslot *hslot2;
352 	struct sock *result, *sk;
353 	unsigned int hash2;
354 
355 	hash2 = ipv6_portaddr_hash(net, daddr, hnum);
356 	hslot2 = udp_hashslot2(udptable, hash2);
357 
358 	if (udp_has_hash4(hslot2)) {
359 		result = udp6_lib_lookup4(net, saddr, sport, daddr, hnum,
360 					  dif, sdif, udptable);
361 		if (result) /* udp6_lib_lookup4 return sk or NULL */
362 			return result;
363 	}
364 
365 	/* Lookup connected or non-wildcard sockets */
366 	result = udp6_lib_lookup2(net, saddr, sport,
367 				  daddr, hnum, dif, sdif,
368 				  hslot2, skb);
369 	if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
370 		goto done;
371 
372 	/* Lookup redirect from BPF */
373 	if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
374 	    udptable == net->ipv4.udp_table) {
375 		sk = inet6_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
376 						saddr, sport, daddr, hnum, dif,
377 						udp6_ehashfn);
378 		if (sk) {
379 			result = sk;
380 			goto done;
381 		}
382 	}
383 
384 	/* Got non-wildcard socket or error on first lookup */
385 	if (result)
386 		goto done;
387 
388 	/* Lookup wildcard sockets */
389 	hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
390 	hslot2 = udp_hashslot2(udptable, hash2);
391 
392 	result = udp6_lib_lookup2(net, saddr, sport,
393 				  &in6addr_any, hnum, dif, sdif,
394 				  hslot2, skb);
395 	if (!IS_ERR_OR_NULL(result))
396 		goto done;
397 
398 	/* Cover address change/lookup/rehash race: see __udp4_lib_lookup() */
399 	result = udp6_lib_lookup1(net, saddr, sport, daddr, hnum, dif, sdif,
400 				  udptable);
401 
402 done:
403 	if (IS_ERR(result))
404 		return NULL;
405 	return result;
406 }
407 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
408 
409 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
410 					  __be16 sport, __be16 dport,
411 					  struct udp_table *udptable)
412 {
413 	const struct ipv6hdr *iph = ipv6_hdr(skb);
414 
415 	return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
416 				 &iph->daddr, dport, inet6_iif(skb),
417 				 inet6_sdif(skb), udptable, skb);
418 }
419 
420 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
421 				 __be16 sport, __be16 dport)
422 {
423 	const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
424 	const struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + offset);
425 	struct net *net = dev_net(skb->dev);
426 	int iif, sdif;
427 
428 	inet6_get_iif_sdif(skb, &iif, &sdif);
429 
430 	return __udp6_lib_lookup(net, &iph->saddr, sport,
431 				 &iph->daddr, dport, iif,
432 				 sdif, net->ipv4.udp_table, NULL);
433 }
434 
435 /* Must be called under rcu_read_lock().
436  * Does increment socket refcount.
437  */
438 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
439 struct sock *udp6_lib_lookup(const struct net *net, const struct in6_addr *saddr, __be16 sport,
440 			     const struct in6_addr *daddr, __be16 dport, int dif)
441 {
442 	struct sock *sk;
443 
444 	sk =  __udp6_lib_lookup(net, saddr, sport, daddr, dport,
445 				dif, 0, net->ipv4.udp_table, NULL);
446 	if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
447 		sk = NULL;
448 	return sk;
449 }
450 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
451 #endif
452 
453 /* do not use the scratch area len for jumbogram: their length exceeds the
454  * scratch area space; note that the IP6CB flags is still in the first
455  * cacheline, so checking for jumbograms is cheap
456  */
457 static int udp6_skb_len(struct sk_buff *skb)
458 {
459 	return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
460 }
461 
462 /*
463  *	This should be easy, if there is something there we
464  *	return it, otherwise we block.
465  */
466 
467 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
468 		  int flags, int *addr_len)
469 {
470 	struct ipv6_pinfo *np = inet6_sk(sk);
471 	struct inet_sock *inet = inet_sk(sk);
472 	struct sk_buff *skb;
473 	unsigned int ulen, copied;
474 	int off, err, peeking = flags & MSG_PEEK;
475 	int is_udplite = IS_UDPLITE(sk);
476 	struct udp_mib __percpu *mib;
477 	bool checksum_valid = false;
478 	int is_udp4;
479 
480 	if (flags & MSG_ERRQUEUE)
481 		return ipv6_recv_error(sk, msg, len, addr_len);
482 
483 	if (np->rxopt.bits.rxpmtu && READ_ONCE(np->rxpmtu))
484 		return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
485 
486 try_again:
487 	off = sk_peek_offset(sk, flags);
488 	skb = __skb_recv_udp(sk, flags, &off, &err);
489 	if (!skb)
490 		return err;
491 
492 	ulen = udp6_skb_len(skb);
493 	copied = len;
494 	if (copied > ulen - off)
495 		copied = ulen - off;
496 	else if (copied < ulen)
497 		msg->msg_flags |= MSG_TRUNC;
498 
499 	is_udp4 = (skb->protocol == htons(ETH_P_IP));
500 	mib = __UDPX_MIB(sk, is_udp4);
501 
502 	/*
503 	 * If checksum is needed at all, try to do it while copying the
504 	 * data.  If the data is truncated, or if we only want a partial
505 	 * coverage checksum (UDP-Lite), do it before the copy.
506 	 */
507 
508 	if (copied < ulen || peeking ||
509 	    (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
510 		checksum_valid = udp_skb_csum_unnecessary(skb) ||
511 				!__udp_lib_checksum_complete(skb);
512 		if (!checksum_valid)
513 			goto csum_copy_err;
514 	}
515 
516 	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
517 		if (udp_skb_is_linear(skb))
518 			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
519 		else
520 			err = skb_copy_datagram_msg(skb, off, msg, copied);
521 	} else {
522 		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
523 		if (err == -EINVAL)
524 			goto csum_copy_err;
525 	}
526 	if (unlikely(err)) {
527 		if (!peeking) {
528 			udp_drops_inc(sk);
529 			SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
530 		}
531 		kfree_skb(skb);
532 		return err;
533 	}
534 	if (!peeking)
535 		SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
536 
537 	sock_recv_cmsgs(msg, sk, skb);
538 
539 	/* Copy the address. */
540 	if (msg->msg_name) {
541 		DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
542 		sin6->sin6_family = AF_INET6;
543 		sin6->sin6_port = udp_hdr(skb)->source;
544 		sin6->sin6_flowinfo = 0;
545 
546 		if (is_udp4) {
547 			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
548 					       &sin6->sin6_addr);
549 			sin6->sin6_scope_id = 0;
550 		} else {
551 			sin6->sin6_addr = ipv6_hdr(skb)->saddr;
552 			sin6->sin6_scope_id =
553 				ipv6_iface_scope_id(&sin6->sin6_addr,
554 						    inet6_iif(skb));
555 		}
556 		*addr_len = sizeof(*sin6);
557 
558 		BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
559 						      (struct sockaddr *)sin6,
560 						      addr_len);
561 	}
562 
563 	if (udp_test_bit(GRO_ENABLED, sk))
564 		udp_cmsg_recv(msg, sk, skb);
565 
566 	if (np->rxopt.all)
567 		ip6_datagram_recv_common_ctl(sk, msg, skb);
568 
569 	if (is_udp4) {
570 		if (inet_cmsg_flags(inet))
571 			ip_cmsg_recv_offset(msg, sk, skb,
572 					    sizeof(struct udphdr), off);
573 	} else {
574 		if (np->rxopt.all)
575 			ip6_datagram_recv_specific_ctl(sk, msg, skb);
576 	}
577 
578 	err = copied;
579 	if (flags & MSG_TRUNC)
580 		err = ulen;
581 
582 	skb_consume_udp(sk, skb, peeking ? -err : err);
583 	return err;
584 
585 csum_copy_err:
586 	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
587 				 udp_skb_destructor)) {
588 		SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS);
589 		SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
590 	}
591 	kfree_skb_reason(skb, SKB_DROP_REASON_UDP_CSUM);
592 
593 	/* starting over for a new packet, but check if we need to yield */
594 	cond_resched();
595 	msg->msg_flags &= ~MSG_TRUNC;
596 	goto try_again;
597 }
598 
599 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
600 void udpv6_encap_enable(void)
601 {
602 	static_branch_inc(&udpv6_encap_needed_key);
603 }
604 EXPORT_SYMBOL(udpv6_encap_enable);
605 
606 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
607  * through error handlers in encapsulations looking for a match.
608  */
609 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb,
610 				      struct inet6_skb_parm *opt,
611 				      u8 type, u8 code, int offset, __be32 info)
612 {
613 	int i;
614 
615 	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
616 		int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt,
617 			       u8 type, u8 code, int offset, __be32 info);
618 		const struct ip6_tnl_encap_ops *encap;
619 
620 		encap = rcu_dereference(ip6tun_encaps[i]);
621 		if (!encap)
622 			continue;
623 		handler = encap->err_handler;
624 		if (handler && !handler(skb, opt, type, code, offset, info))
625 			return 0;
626 	}
627 
628 	return -ENOENT;
629 }
630 
631 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
632  * reversing source and destination port: this will match tunnels that force the
633  * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
634  * lwtunnels might actually break this assumption by being configured with
635  * different destination ports on endpoints, in this case we won't be able to
636  * trace ICMP messages back to them.
637  *
638  * If this doesn't match any socket, probe tunnels with arbitrary destination
639  * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
640  * we've sent packets to won't necessarily match the local destination port.
641  *
642  * Then ask the tunnel implementation to match the error against a valid
643  * association.
644  *
645  * Return an error if we can't find a match, the socket if we need further
646  * processing, zero otherwise.
647  */
648 static struct sock *__udp6_lib_err_encap(struct net *net,
649 					 const struct ipv6hdr *hdr, int offset,
650 					 struct udphdr *uh,
651 					 struct udp_table *udptable,
652 					 struct sock *sk,
653 					 struct sk_buff *skb,
654 					 struct inet6_skb_parm *opt,
655 					 u8 type, u8 code, __be32 info)
656 {
657 	int (*lookup)(struct sock *sk, struct sk_buff *skb);
658 	int network_offset, transport_offset;
659 	struct udp_sock *up;
660 
661 	network_offset = skb_network_offset(skb);
662 	transport_offset = skb_transport_offset(skb);
663 
664 	/* Network header needs to point to the outer IPv6 header inside ICMP */
665 	skb_reset_network_header(skb);
666 
667 	/* Transport header needs to point to the UDP header */
668 	skb_set_transport_header(skb, offset);
669 
670 	if (sk) {
671 		up = udp_sk(sk);
672 
673 		lookup = READ_ONCE(up->encap_err_lookup);
674 		if (lookup && lookup(sk, skb))
675 			sk = NULL;
676 
677 		goto out;
678 	}
679 
680 	sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
681 			       &hdr->saddr, uh->dest,
682 			       inet6_iif(skb), 0, udptable, skb);
683 	if (sk) {
684 		up = udp_sk(sk);
685 
686 		lookup = READ_ONCE(up->encap_err_lookup);
687 		if (!lookup || lookup(sk, skb))
688 			sk = NULL;
689 	}
690 
691 out:
692 	if (!sk) {
693 		sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
694 							offset, info));
695 	}
696 
697 	skb_set_transport_header(skb, transport_offset);
698 	skb_set_network_header(skb, network_offset);
699 
700 	return sk;
701 }
702 
703 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
704 		   u8 type, u8 code, int offset, __be32 info,
705 		   struct udp_table *udptable)
706 {
707 	struct ipv6_pinfo *np;
708 	const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
709 	const struct in6_addr *saddr = &hdr->saddr;
710 	const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr;
711 	struct udphdr *uh = (struct udphdr *)(skb->data+offset);
712 	bool tunnel = false;
713 	struct sock *sk;
714 	int harderr;
715 	int err;
716 	struct net *net = dev_net(skb->dev);
717 
718 	sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
719 			       inet6_iif(skb), inet6_sdif(skb), udptable, NULL);
720 
721 	if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
722 		/* No socket for error: try tunnels before discarding */
723 		if (static_branch_unlikely(&udpv6_encap_needed_key)) {
724 			sk = __udp6_lib_err_encap(net, hdr, offset, uh,
725 						  udptable, sk, skb,
726 						  opt, type, code, info);
727 			if (!sk)
728 				return 0;
729 		} else
730 			sk = ERR_PTR(-ENOENT);
731 
732 		if (IS_ERR(sk)) {
733 			__ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
734 					  ICMP6_MIB_INERRORS);
735 			return PTR_ERR(sk);
736 		}
737 
738 		tunnel = true;
739 	}
740 
741 	harderr = icmpv6_err_convert(type, code, &err);
742 	np = inet6_sk(sk);
743 
744 	if (type == ICMPV6_PKT_TOOBIG) {
745 		if (!ip6_sk_accept_pmtu(sk))
746 			goto out;
747 		ip6_sk_update_pmtu(skb, sk, info);
748 		if (READ_ONCE(np->pmtudisc) != IPV6_PMTUDISC_DONT)
749 			harderr = 1;
750 	}
751 	if (type == NDISC_REDIRECT) {
752 		if (tunnel) {
753 			ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
754 				     READ_ONCE(sk->sk_mark),
755 				     sk_uid(sk));
756 		} else {
757 			ip6_sk_redirect(skb, sk);
758 		}
759 		goto out;
760 	}
761 
762 	/* Tunnels don't have an application socket: don't pass errors back */
763 	if (tunnel) {
764 		if (udp_sk(sk)->encap_err_rcv)
765 			udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest,
766 						  ntohl(info), (u8 *)(uh+1));
767 		goto out;
768 	}
769 
770 	if (!inet6_test_bit(RECVERR6, sk)) {
771 		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
772 			goto out;
773 	} else {
774 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
775 	}
776 
777 	sk->sk_err = err;
778 	sk_error_report(sk);
779 out:
780 	return 0;
781 }
782 
783 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
784 {
785 	int rc;
786 
787 	if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
788 		sock_rps_save_rxhash(sk, skb);
789 		sk_mark_napi_id(sk, skb);
790 		sk_incoming_cpu_update(sk);
791 	} else {
792 		sk_mark_napi_id_once(sk, skb);
793 	}
794 
795 	rc = __udp_enqueue_schedule_skb(sk, skb);
796 	if (rc < 0) {
797 		int is_udplite = IS_UDPLITE(sk);
798 		enum skb_drop_reason drop_reason;
799 
800 		/* Note that an ENOMEM error is charged twice */
801 		if (rc == -ENOMEM) {
802 			UDP6_INC_STATS(sock_net(sk),
803 					 UDP_MIB_RCVBUFERRORS, is_udplite);
804 			drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
805 		} else {
806 			UDP6_INC_STATS(sock_net(sk),
807 				       UDP_MIB_MEMERRORS, is_udplite);
808 			drop_reason = SKB_DROP_REASON_PROTO_MEM;
809 		}
810 		UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
811 		trace_udp_fail_queue_rcv_skb(rc, sk, skb);
812 		sk_skb_reason_drop(sk, skb, drop_reason);
813 		return -1;
814 	}
815 
816 	return 0;
817 }
818 
819 static __inline__ int udpv6_err(struct sk_buff *skb,
820 				struct inet6_skb_parm *opt, u8 type,
821 				u8 code, int offset, __be32 info)
822 {
823 	return __udp6_lib_err(skb, opt, type, code, offset, info,
824 			      dev_net(skb->dev)->ipv4.udp_table);
825 }
826 
827 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
828 {
829 	enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
830 	struct udp_sock *up = udp_sk(sk);
831 	int is_udplite = IS_UDPLITE(sk);
832 
833 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
834 		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
835 		goto drop;
836 	}
837 	nf_reset_ct(skb);
838 
839 	if (static_branch_unlikely(&udpv6_encap_needed_key) &&
840 	    READ_ONCE(up->encap_type)) {
841 		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
842 
843 		/*
844 		 * This is an encapsulation socket so pass the skb to
845 		 * the socket's udp_encap_rcv() hook. Otherwise, just
846 		 * fall through and pass this up the UDP socket.
847 		 * up->encap_rcv() returns the following value:
848 		 * =0 if skb was successfully passed to the encap
849 		 *    handler or was discarded by it.
850 		 * >0 if skb should be passed on to UDP.
851 		 * <0 if skb should be resubmitted as proto -N
852 		 */
853 
854 		/* if we're overly short, let UDP handle it */
855 		encap_rcv = READ_ONCE(up->encap_rcv);
856 		if (encap_rcv) {
857 			int ret;
858 
859 			/* Verify checksum before giving to encap */
860 			if (udp_lib_checksum_complete(skb))
861 				goto csum_error;
862 
863 			ret = encap_rcv(sk, skb);
864 			if (ret <= 0) {
865 				__UDP6_INC_STATS(sock_net(sk),
866 						 UDP_MIB_INDATAGRAMS,
867 						 is_udplite);
868 				return -ret;
869 			}
870 		}
871 
872 		/* FALLTHROUGH -- it's a UDP Packet */
873 	}
874 
875 	/*
876 	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
877 	 */
878 	if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
879 		u16 pcrlen = READ_ONCE(up->pcrlen);
880 
881 		if (pcrlen == 0) {          /* full coverage was set  */
882 			net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
883 					    UDP_SKB_CB(skb)->cscov, skb->len);
884 			goto drop;
885 		}
886 		if (UDP_SKB_CB(skb)->cscov < pcrlen) {
887 			net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
888 					    UDP_SKB_CB(skb)->cscov, pcrlen);
889 			goto drop;
890 		}
891 	}
892 
893 	prefetch(&sk->sk_rmem_alloc);
894 	if (rcu_access_pointer(sk->sk_filter) &&
895 	    udp_lib_checksum_complete(skb))
896 		goto csum_error;
897 
898 	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr), &drop_reason))
899 		goto drop;
900 
901 	udp_csum_pull_header(skb);
902 
903 	skb_dst_drop(skb);
904 
905 	return __udpv6_queue_rcv_skb(sk, skb);
906 
907 csum_error:
908 	drop_reason = SKB_DROP_REASON_UDP_CSUM;
909 	__UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
910 drop:
911 	__UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
912 	udp_drops_inc(sk);
913 	sk_skb_reason_drop(sk, skb, drop_reason);
914 	return -1;
915 }
916 
917 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
918 {
919 	struct sk_buff *next, *segs;
920 	int ret;
921 
922 	if (likely(!udp_unexpected_gso(sk, skb)))
923 		return udpv6_queue_rcv_one_skb(sk, skb);
924 
925 	__skb_push(skb, -skb_mac_offset(skb));
926 	segs = udp_rcv_segment(sk, skb, false);
927 	skb_list_walk_safe(segs, skb, next) {
928 		__skb_pull(skb, skb_transport_offset(skb));
929 
930 		udp_post_segment_fix_csum(skb);
931 		ret = udpv6_queue_rcv_one_skb(sk, skb);
932 		if (ret > 0)
933 			ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
934 						 true);
935 	}
936 	return 0;
937 }
938 
939 static bool __udp_v6_is_mcast_sock(struct net *net, const struct sock *sk,
940 				   __be16 loc_port, const struct in6_addr *loc_addr,
941 				   __be16 rmt_port, const struct in6_addr *rmt_addr,
942 				   int dif, int sdif, unsigned short hnum)
943 {
944 	const struct inet_sock *inet = inet_sk(sk);
945 
946 	if (!net_eq(sock_net(sk), net))
947 		return false;
948 
949 	if (udp_sk(sk)->udp_port_hash != hnum ||
950 	    sk->sk_family != PF_INET6 ||
951 	    (inet->inet_dport && inet->inet_dport != rmt_port) ||
952 	    (!ipv6_addr_any(&sk->sk_v6_daddr) &&
953 		    !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
954 	    !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) ||
955 	    (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
956 		    !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
957 		return false;
958 	if (!inet6_mc_check(sk, loc_addr, rmt_addr))
959 		return false;
960 	return true;
961 }
962 
963 static void udp6_csum_zero_error(struct sk_buff *skb)
964 {
965 	/* RFC 2460 section 8.1 says that we SHOULD log
966 	 * this error. Well, it is reasonable.
967 	 */
968 	net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
969 			    &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
970 			    &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
971 }
972 
973 /*
974  * Note: called only from the BH handler context,
975  * so we don't need to lock the hashes.
976  */
977 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
978 		const struct in6_addr *saddr, const struct in6_addr *daddr,
979 		struct udp_table *udptable, int proto)
980 {
981 	struct sock *sk, *first = NULL;
982 	const struct udphdr *uh = udp_hdr(skb);
983 	unsigned short hnum = ntohs(uh->dest);
984 	struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
985 	unsigned int offset = offsetof(typeof(*sk), sk_node);
986 	unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
987 	int dif = inet6_iif(skb);
988 	int sdif = inet6_sdif(skb);
989 	struct hlist_node *node;
990 	struct sk_buff *nskb;
991 
992 	if (use_hash2) {
993 		hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
994 			    udptable->mask;
995 		hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
996 start_lookup:
997 		hslot = &udptable->hash2[hash2].hslot;
998 		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
999 	}
1000 
1001 	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1002 		if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
1003 					    uh->source, saddr, dif, sdif,
1004 					    hnum))
1005 			continue;
1006 		/* If zero checksum and no_check is not on for
1007 		 * the socket then skip it.
1008 		 */
1009 		if (!uh->check && !udp_get_no_check6_rx(sk))
1010 			continue;
1011 		if (!first) {
1012 			first = sk;
1013 			continue;
1014 		}
1015 		nskb = skb_clone(skb, GFP_ATOMIC);
1016 		if (unlikely(!nskb)) {
1017 			udp_drops_inc(sk);
1018 			__UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1019 					 IS_UDPLITE(sk));
1020 			__UDP6_INC_STATS(net, UDP_MIB_INERRORS,
1021 					 IS_UDPLITE(sk));
1022 			continue;
1023 		}
1024 
1025 		if (udpv6_queue_rcv_skb(sk, nskb) > 0)
1026 			consume_skb(nskb);
1027 	}
1028 
1029 	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
1030 	if (use_hash2 && hash2 != hash2_any) {
1031 		hash2 = hash2_any;
1032 		goto start_lookup;
1033 	}
1034 
1035 	if (first) {
1036 		if (udpv6_queue_rcv_skb(first, skb) > 0)
1037 			consume_skb(skb);
1038 	} else {
1039 		kfree_skb(skb);
1040 		__UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
1041 				 proto == IPPROTO_UDPLITE);
1042 	}
1043 	return 0;
1044 }
1045 
1046 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1047 {
1048 	if (udp_sk_rx_dst_set(sk, dst))
1049 		sk->sk_rx_dst_cookie = rt6_get_cookie(dst_rt6_info(dst));
1050 }
1051 
1052 /* wrapper for udp_queue_rcv_skb taking care of csum conversion and
1053  * return code conversion for ip layer consumption
1054  */
1055 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
1056 				struct udphdr *uh)
1057 {
1058 	int ret;
1059 
1060 	if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1061 		skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo);
1062 
1063 	ret = udpv6_queue_rcv_skb(sk, skb);
1064 
1065 	/* a return value > 0 means to resubmit the input */
1066 	if (ret > 0)
1067 		return ret;
1068 	return 0;
1069 }
1070 
1071 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1072 		   int proto)
1073 {
1074 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1075 	const struct in6_addr *saddr, *daddr;
1076 	struct net *net = dev_net(skb->dev);
1077 	struct sock *sk = NULL;
1078 	struct udphdr *uh;
1079 	bool refcounted;
1080 	u32 ulen = 0;
1081 
1082 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1083 		goto discard;
1084 
1085 	saddr = &ipv6_hdr(skb)->saddr;
1086 	daddr = &ipv6_hdr(skb)->daddr;
1087 	uh = udp_hdr(skb);
1088 
1089 	ulen = ntohs(uh->len);
1090 	if (ulen > skb->len)
1091 		goto short_packet;
1092 
1093 	if (proto == IPPROTO_UDP) {
1094 		/* UDP validates ulen. */
1095 
1096 		/* Check for jumbo payload */
1097 		if (ulen == 0)
1098 			ulen = skb->len;
1099 
1100 		if (ulen < sizeof(*uh))
1101 			goto short_packet;
1102 
1103 		if (ulen < skb->len) {
1104 			if (pskb_trim_rcsum(skb, ulen))
1105 				goto short_packet;
1106 			saddr = &ipv6_hdr(skb)->saddr;
1107 			daddr = &ipv6_hdr(skb)->daddr;
1108 			uh = udp_hdr(skb);
1109 		}
1110 	}
1111 
1112 	if (udp6_csum_init(skb, uh, proto))
1113 		goto csum_error;
1114 
1115 	/* Check if the socket is already available, e.g. due to early demux */
1116 	sk = inet6_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
1117 			      &refcounted, udp6_ehashfn);
1118 	if (IS_ERR(sk))
1119 		goto no_sk;
1120 
1121 	if (sk) {
1122 		struct dst_entry *dst = skb_dst(skb);
1123 		int ret;
1124 
1125 		if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
1126 			udp6_sk_rx_dst_set(sk, dst);
1127 
1128 		if (!uh->check && !udp_get_no_check6_rx(sk)) {
1129 			if (refcounted)
1130 				sock_put(sk);
1131 			goto report_csum_error;
1132 		}
1133 
1134 		ret = udp6_unicast_rcv_skb(sk, skb, uh);
1135 		if (refcounted)
1136 			sock_put(sk);
1137 		return ret;
1138 	}
1139 
1140 	/*
1141 	 *	Multicast receive code
1142 	 */
1143 	if (ipv6_addr_is_multicast(daddr))
1144 		return __udp6_lib_mcast_deliver(net, skb,
1145 				saddr, daddr, udptable, proto);
1146 
1147 	/* Unicast */
1148 	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1149 	if (sk) {
1150 		if (!uh->check && !udp_get_no_check6_rx(sk))
1151 			goto report_csum_error;
1152 		return udp6_unicast_rcv_skb(sk, skb, uh);
1153 	}
1154 no_sk:
1155 	reason = SKB_DROP_REASON_NO_SOCKET;
1156 
1157 	if (!uh->check)
1158 		goto report_csum_error;
1159 
1160 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1161 		goto discard;
1162 	nf_reset_ct(skb);
1163 
1164 	if (udp_lib_checksum_complete(skb))
1165 		goto csum_error;
1166 
1167 	__UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1168 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
1169 
1170 	sk_skb_reason_drop(sk, skb, reason);
1171 	return 0;
1172 
1173 short_packet:
1174 	if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1175 		reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1176 	net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
1177 			    proto == IPPROTO_UDPLITE ? "-Lite" : "",
1178 			    saddr, ntohs(uh->source),
1179 			    ulen, skb->len,
1180 			    daddr, ntohs(uh->dest));
1181 	goto discard;
1182 
1183 report_csum_error:
1184 	udp6_csum_zero_error(skb);
1185 csum_error:
1186 	if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1187 		reason = SKB_DROP_REASON_UDP_CSUM;
1188 	__UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1189 discard:
1190 	__UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1191 	sk_skb_reason_drop(sk, skb, reason);
1192 	return 0;
1193 }
1194 
1195 
1196 static struct sock *__udp6_lib_demux_lookup(struct net *net,
1197 			__be16 loc_port, const struct in6_addr *loc_addr,
1198 			__be16 rmt_port, const struct in6_addr *rmt_addr,
1199 			int dif, int sdif)
1200 {
1201 	struct udp_table *udptable = net->ipv4.udp_table;
1202 	unsigned short hnum = ntohs(loc_port);
1203 	struct udp_hslot *hslot2;
1204 	unsigned int hash2;
1205 	__portpair ports;
1206 	struct sock *sk;
1207 
1208 	hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
1209 	hslot2 = udp_hashslot2(udptable, hash2);
1210 	ports = INET_COMBINED_PORTS(rmt_port, hnum);
1211 
1212 	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1213 		if (sk->sk_state == TCP_ESTABLISHED &&
1214 		    inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif))
1215 			return sk;
1216 		/* Only check first socket in chain */
1217 		break;
1218 	}
1219 	return NULL;
1220 }
1221 
1222 void udp_v6_early_demux(struct sk_buff *skb)
1223 {
1224 	struct net *net = dev_net(skb->dev);
1225 	const struct udphdr *uh;
1226 	struct sock *sk;
1227 	struct dst_entry *dst;
1228 	int dif = skb->dev->ifindex;
1229 	int sdif = inet6_sdif(skb);
1230 
1231 	if (!pskb_may_pull(skb, skb_transport_offset(skb) +
1232 	    sizeof(struct udphdr)))
1233 		return;
1234 
1235 	uh = udp_hdr(skb);
1236 
1237 	if (skb->pkt_type == PACKET_HOST)
1238 		sk = __udp6_lib_demux_lookup(net, uh->dest,
1239 					     &ipv6_hdr(skb)->daddr,
1240 					     uh->source, &ipv6_hdr(skb)->saddr,
1241 					     dif, sdif);
1242 	else
1243 		return;
1244 
1245 	if (!sk)
1246 		return;
1247 
1248 	skb->sk = sk;
1249 	DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
1250 	skb->destructor = sock_pfree;
1251 	dst = rcu_dereference(sk->sk_rx_dst);
1252 
1253 	if (dst)
1254 		dst = dst_check(dst, sk->sk_rx_dst_cookie);
1255 	if (dst) {
1256 		/* set noref for now.
1257 		 * any place which wants to hold dst has to call
1258 		 * dst_hold_safe()
1259 		 */
1260 		skb_dst_set_noref(skb, dst);
1261 	}
1262 }
1263 
1264 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
1265 {
1266 	return __udp6_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
1267 }
1268 
1269 /*
1270  * Throw away all pending data and cancel the corking. Socket is locked.
1271  */
1272 static void udp_v6_flush_pending_frames(struct sock *sk)
1273 {
1274 	struct udp_sock *up = udp_sk(sk);
1275 
1276 	if (up->pending == AF_INET)
1277 		udp_flush_pending_frames(sk);
1278 	else if (up->pending) {
1279 		up->len = 0;
1280 		WRITE_ONCE(up->pending, 0);
1281 		ip6_flush_pending_frames(sk);
1282 	}
1283 }
1284 
1285 static int udpv6_pre_connect(struct sock *sk, struct sockaddr_unsized *uaddr,
1286 			     int addr_len)
1287 {
1288 	if (addr_len < offsetofend(struct sockaddr, sa_family))
1289 		return -EINVAL;
1290 	/* The following checks are replicated from __ip6_datagram_connect()
1291 	 * and intended to prevent BPF program called below from accessing
1292 	 * bytes that are out of the bound specified by user in addr_len.
1293 	 */
1294 	if (uaddr->sa_family == AF_INET) {
1295 		if (ipv6_only_sock(sk))
1296 			return -EAFNOSUPPORT;
1297 		return udp_pre_connect(sk, uaddr, addr_len);
1298 	}
1299 
1300 	if (addr_len < SIN6_LEN_RFC2133)
1301 		return -EINVAL;
1302 
1303 	return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, &addr_len);
1304 }
1305 
1306 static int udpv6_connect(struct sock *sk, struct sockaddr_unsized *uaddr,
1307 			 int addr_len)
1308 {
1309 	int res;
1310 
1311 	lock_sock(sk);
1312 	res = __ip6_datagram_connect(sk, uaddr, addr_len);
1313 	if (!res)
1314 		udp6_hash4(sk);
1315 	release_sock(sk);
1316 	return res;
1317 }
1318 
1319 /**
1320  *	udp6_hwcsum_outgoing  -  handle outgoing HW checksumming
1321  *	@sk:	socket we are sending on
1322  *	@skb:	sk_buff containing the filled-in UDP header
1323  *		(checksum field must be zeroed out)
1324  *	@saddr: source address
1325  *	@daddr: destination address
1326  *	@len:	length of packet
1327  */
1328 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1329 				 const struct in6_addr *saddr,
1330 				 const struct in6_addr *daddr, int len)
1331 {
1332 	unsigned int offset;
1333 	struct udphdr *uh = udp_hdr(skb);
1334 	struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1335 	__wsum csum = 0;
1336 
1337 	if (!frags) {
1338 		/* Only one fragment on the socket.  */
1339 		skb->csum_start = skb_transport_header(skb) - skb->head;
1340 		skb->csum_offset = offsetof(struct udphdr, check);
1341 		uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1342 	} else {
1343 		/*
1344 		 * HW-checksum won't work as there are two or more
1345 		 * fragments on the socket so that all csums of sk_buffs
1346 		 * should be together
1347 		 */
1348 		offset = skb_transport_offset(skb);
1349 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1350 		csum = skb->csum;
1351 
1352 		skb->ip_summed = CHECKSUM_NONE;
1353 
1354 		do {
1355 			csum = csum_add(csum, frags->csum);
1356 		} while ((frags = frags->next));
1357 
1358 		uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1359 					    csum);
1360 		if (uh->check == 0)
1361 			uh->check = CSUM_MANGLED_0;
1362 	}
1363 }
1364 
1365 /*
1366  *	Sending
1367  */
1368 
1369 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1370 			   struct inet_cork *cork)
1371 {
1372 	struct sock *sk = skb->sk;
1373 	struct udphdr *uh;
1374 	int err = 0;
1375 	int is_udplite = IS_UDPLITE(sk);
1376 	__wsum csum = 0;
1377 	int offset = skb_transport_offset(skb);
1378 	int len = skb->len - offset;
1379 	int datalen = len - sizeof(*uh);
1380 
1381 	/*
1382 	 * Create a UDP header
1383 	 */
1384 	uh = udp_hdr(skb);
1385 	uh->source = fl6->fl6_sport;
1386 	uh->dest = fl6->fl6_dport;
1387 	uh->len = htons(len);
1388 	uh->check = 0;
1389 
1390 	if (cork->gso_size) {
1391 		const int hlen = skb_network_header_len(skb) +
1392 				 sizeof(struct udphdr);
1393 
1394 		if (hlen + min(datalen, cork->gso_size) > cork->fragsize) {
1395 			kfree_skb(skb);
1396 			return -EMSGSIZE;
1397 		}
1398 		if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1399 			kfree_skb(skb);
1400 			return -EINVAL;
1401 		}
1402 		if (udp_get_no_check6_tx(sk)) {
1403 			kfree_skb(skb);
1404 			return -EINVAL;
1405 		}
1406 		if (is_udplite || dst_xfrm(skb_dst(skb))) {
1407 			kfree_skb(skb);
1408 			return -EIO;
1409 		}
1410 
1411 		if (datalen > cork->gso_size) {
1412 			skb_shinfo(skb)->gso_size = cork->gso_size;
1413 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1414 			skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1415 								 cork->gso_size);
1416 
1417 			/* Don't checksum the payload, skb will get segmented */
1418 			goto csum_partial;
1419 		}
1420 	}
1421 
1422 	if (is_udplite)
1423 		csum = udplite_csum(skb);
1424 	else if (udp_get_no_check6_tx(sk)) {   /* UDP csum disabled */
1425 		skb->ip_summed = CHECKSUM_NONE;
1426 		goto send;
1427 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1428 csum_partial:
1429 		udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1430 		goto send;
1431 	} else
1432 		csum = udp_csum(skb);
1433 
1434 	/* add protocol-dependent pseudo-header */
1435 	uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1436 				    len, fl6->flowi6_proto, csum);
1437 	if (uh->check == 0)
1438 		uh->check = CSUM_MANGLED_0;
1439 
1440 send:
1441 	err = ip6_send_skb(skb);
1442 	if (err) {
1443 		if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk)) {
1444 			UDP6_INC_STATS(sock_net(sk),
1445 				       UDP_MIB_SNDBUFERRORS, is_udplite);
1446 			err = 0;
1447 		}
1448 	} else {
1449 		UDP6_INC_STATS(sock_net(sk),
1450 			       UDP_MIB_OUTDATAGRAMS, is_udplite);
1451 	}
1452 	return err;
1453 }
1454 
1455 static int udp_v6_push_pending_frames(struct sock *sk)
1456 {
1457 	struct sk_buff *skb;
1458 	struct udp_sock  *up = udp_sk(sk);
1459 	int err = 0;
1460 
1461 	if (up->pending == AF_INET)
1462 		return udp_push_pending_frames(sk);
1463 
1464 	skb = ip6_finish_skb(sk);
1465 	if (!skb)
1466 		goto out;
1467 
1468 	err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6,
1469 			      &inet_sk(sk)->cork.base);
1470 out:
1471 	up->len = 0;
1472 	WRITE_ONCE(up->pending, 0);
1473 	return err;
1474 }
1475 
1476 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1477 {
1478 	struct ipv6_txoptions opt_space;
1479 	struct udp_sock *up = udp_sk(sk);
1480 	struct inet_sock *inet = inet_sk(sk);
1481 	struct ipv6_pinfo *np = inet6_sk(sk);
1482 	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1483 	struct in6_addr *daddr, *final_p, final;
1484 	struct ipv6_txoptions *opt = NULL;
1485 	struct ipv6_txoptions *opt_to_free = NULL;
1486 	struct ip6_flowlabel *flowlabel = NULL;
1487 	struct inet_cork_full cork;
1488 	struct flowi6 *fl6 = &cork.fl.u.ip6;
1489 	struct dst_entry *dst;
1490 	struct ipcm6_cookie ipc6;
1491 	int addr_len = msg->msg_namelen;
1492 	bool connected = false;
1493 	int ulen = len;
1494 	int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1495 	int err;
1496 	int is_udplite = IS_UDPLITE(sk);
1497 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1498 
1499 	ipcm6_init_sk(&ipc6, sk);
1500 	ipc6.gso_size = READ_ONCE(up->gso_size);
1501 
1502 	/* destination address check */
1503 	if (sin6) {
1504 		if (addr_len < offsetof(struct sockaddr, sa_data))
1505 			return -EINVAL;
1506 
1507 		switch (sin6->sin6_family) {
1508 		case AF_INET6:
1509 			if (addr_len < SIN6_LEN_RFC2133)
1510 				return -EINVAL;
1511 			daddr = &sin6->sin6_addr;
1512 			if (ipv6_addr_any(daddr) &&
1513 			    ipv6_addr_v4mapped(&np->saddr))
1514 				ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1515 						       daddr);
1516 			break;
1517 		case AF_INET:
1518 			goto do_udp_sendmsg;
1519 		case AF_UNSPEC:
1520 			msg->msg_name = sin6 = NULL;
1521 			msg->msg_namelen = addr_len = 0;
1522 			daddr = NULL;
1523 			break;
1524 		default:
1525 			return -EINVAL;
1526 		}
1527 	} else if (!READ_ONCE(up->pending)) {
1528 		if (sk->sk_state != TCP_ESTABLISHED)
1529 			return -EDESTADDRREQ;
1530 		daddr = &sk->sk_v6_daddr;
1531 	} else
1532 		daddr = NULL;
1533 
1534 	if (daddr) {
1535 		if (ipv6_addr_v4mapped(daddr)) {
1536 			struct sockaddr_in sin;
1537 			sin.sin_family = AF_INET;
1538 			sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1539 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
1540 			msg->msg_name = &sin;
1541 			msg->msg_namelen = sizeof(sin);
1542 do_udp_sendmsg:
1543 			err = ipv6_only_sock(sk) ?
1544 				-ENETUNREACH : udp_sendmsg(sk, msg, len);
1545 			msg->msg_name = sin6;
1546 			msg->msg_namelen = addr_len;
1547 			return err;
1548 		}
1549 	}
1550 
1551 	/* Rough check on arithmetic overflow,
1552 	   better check is made in ip6_append_data().
1553 	   */
1554 	if (len > INT_MAX - sizeof(struct udphdr))
1555 		return -EMSGSIZE;
1556 
1557 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
1558 	if (READ_ONCE(up->pending)) {
1559 		if (READ_ONCE(up->pending) == AF_INET)
1560 			return udp_sendmsg(sk, msg, len);
1561 		/*
1562 		 * There are pending frames.
1563 		 * The socket lock must be held while it's corked.
1564 		 */
1565 		lock_sock(sk);
1566 		if (likely(up->pending)) {
1567 			if (unlikely(up->pending != AF_INET6)) {
1568 				release_sock(sk);
1569 				return -EAFNOSUPPORT;
1570 			}
1571 			dst = NULL;
1572 			goto do_append_data;
1573 		}
1574 		release_sock(sk);
1575 	}
1576 	ulen += sizeof(struct udphdr);
1577 
1578 	memset(fl6, 0, sizeof(*fl6));
1579 
1580 	if (sin6) {
1581 		if (sin6->sin6_port == 0)
1582 			return -EINVAL;
1583 
1584 		fl6->fl6_dport = sin6->sin6_port;
1585 		daddr = &sin6->sin6_addr;
1586 
1587 		if (inet6_test_bit(SNDFLOW, sk)) {
1588 			fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1589 			if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) {
1590 				flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1591 				if (IS_ERR(flowlabel))
1592 					return -EINVAL;
1593 			}
1594 		}
1595 
1596 		/*
1597 		 * Otherwise it will be difficult to maintain
1598 		 * sk->sk_dst_cache.
1599 		 */
1600 		if (sk->sk_state == TCP_ESTABLISHED &&
1601 		    ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1602 			daddr = &sk->sk_v6_daddr;
1603 
1604 		if (addr_len >= sizeof(struct sockaddr_in6) &&
1605 		    sin6->sin6_scope_id &&
1606 		    __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1607 			fl6->flowi6_oif = sin6->sin6_scope_id;
1608 	} else {
1609 		if (sk->sk_state != TCP_ESTABLISHED)
1610 			return -EDESTADDRREQ;
1611 
1612 		fl6->fl6_dport = inet->inet_dport;
1613 		daddr = &sk->sk_v6_daddr;
1614 		fl6->flowlabel = np->flow_label;
1615 		connected = true;
1616 	}
1617 
1618 	if (!fl6->flowi6_oif)
1619 		fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if);
1620 
1621 	if (!fl6->flowi6_oif)
1622 		fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1623 
1624 	fl6->flowi6_uid = sk_uid(sk);
1625 
1626 	if (msg->msg_controllen) {
1627 		opt = &opt_space;
1628 		memset(opt, 0, sizeof(struct ipv6_txoptions));
1629 		opt->tot_len = sizeof(*opt);
1630 		ipc6.opt = opt;
1631 
1632 		err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1633 		if (err > 0) {
1634 			err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6,
1635 						    &ipc6);
1636 			connected = false;
1637 		}
1638 		if (err < 0) {
1639 			fl6_sock_release(flowlabel);
1640 			return err;
1641 		}
1642 		if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1643 			flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1644 			if (IS_ERR(flowlabel))
1645 				return -EINVAL;
1646 		}
1647 		if (!(opt->opt_nflen|opt->opt_flen))
1648 			opt = NULL;
1649 	}
1650 	if (!opt) {
1651 		opt = txopt_get(np);
1652 		opt_to_free = opt;
1653 	}
1654 	if (flowlabel)
1655 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
1656 	opt = ipv6_fixup_options(&opt_space, opt);
1657 	ipc6.opt = opt;
1658 
1659 	fl6->flowi6_proto = sk->sk_protocol;
1660 	fl6->flowi6_mark = ipc6.sockc.mark;
1661 	fl6->daddr = *daddr;
1662 	if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr))
1663 		fl6->saddr = np->saddr;
1664 	fl6->fl6_sport = inet->inet_sport;
1665 
1666 	if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) {
1667 		err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1668 					   (struct sockaddr *)sin6,
1669 					   &addr_len,
1670 					   &fl6->saddr);
1671 		if (err)
1672 			goto out_no_dst;
1673 		if (sin6) {
1674 			if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1675 				/* BPF program rewrote IPv6-only by IPv4-mapped
1676 				 * IPv6. It's currently unsupported.
1677 				 */
1678 				err = -ENOTSUPP;
1679 				goto out_no_dst;
1680 			}
1681 			if (sin6->sin6_port == 0) {
1682 				/* BPF program set invalid port. Reject it. */
1683 				err = -EINVAL;
1684 				goto out_no_dst;
1685 			}
1686 			fl6->fl6_dport = sin6->sin6_port;
1687 			fl6->daddr = sin6->sin6_addr;
1688 		}
1689 	}
1690 
1691 	if (ipv6_addr_any(&fl6->daddr))
1692 		fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1693 
1694 	final_p = fl6_update_dst(fl6, opt, &final);
1695 	if (final_p)
1696 		connected = false;
1697 
1698 	if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) {
1699 		fl6->flowi6_oif = READ_ONCE(np->mcast_oif);
1700 		connected = false;
1701 	} else if (!fl6->flowi6_oif)
1702 		fl6->flowi6_oif = READ_ONCE(np->ucast_oif);
1703 
1704 	security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6));
1705 
1706 	fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel);
1707 
1708 	dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected);
1709 	if (IS_ERR(dst)) {
1710 		err = PTR_ERR(dst);
1711 		dst = NULL;
1712 		goto out;
1713 	}
1714 
1715 	if (ipc6.hlimit < 0)
1716 		ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst);
1717 
1718 	if (msg->msg_flags&MSG_CONFIRM)
1719 		goto do_confirm;
1720 back_from_confirm:
1721 
1722 	/* Lockless fast path for the non-corking case */
1723 	if (!corkreq) {
1724 		struct sk_buff *skb;
1725 
1726 		skb = ip6_make_skb(sk, getfrag, msg, ulen,
1727 				   sizeof(struct udphdr), &ipc6,
1728 				   dst_rt6_info(dst),
1729 				   msg->msg_flags, &cork);
1730 		err = PTR_ERR(skb);
1731 		if (!IS_ERR_OR_NULL(skb))
1732 			err = udp_v6_send_skb(skb, fl6, &cork.base);
1733 		/* ip6_make_skb steals dst reference */
1734 		goto out_no_dst;
1735 	}
1736 
1737 	lock_sock(sk);
1738 	if (unlikely(up->pending)) {
1739 		/* The socket is already corked while preparing it. */
1740 		/* ... which is an evident application bug. --ANK */
1741 		release_sock(sk);
1742 
1743 		net_dbg_ratelimited("udp cork app bug 2\n");
1744 		err = -EINVAL;
1745 		goto out;
1746 	}
1747 
1748 	WRITE_ONCE(up->pending, AF_INET6);
1749 
1750 do_append_data:
1751 	up->len += ulen;
1752 	err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1753 			      &ipc6, fl6, dst_rt6_info(dst),
1754 			      corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1755 	if (err)
1756 		udp_v6_flush_pending_frames(sk);
1757 	else if (!corkreq)
1758 		err = udp_v6_push_pending_frames(sk);
1759 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1760 		WRITE_ONCE(up->pending, 0);
1761 
1762 	if (err > 0)
1763 		err = inet6_test_bit(RECVERR6, sk) ? net_xmit_errno(err) : 0;
1764 	release_sock(sk);
1765 
1766 out:
1767 	dst_release(dst);
1768 out_no_dst:
1769 	fl6_sock_release(flowlabel);
1770 	txopt_put(opt_to_free);
1771 	if (!err)
1772 		return len;
1773 	/*
1774 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1775 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1776 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1777 	 * things).  We could add another new stat but at least for now that
1778 	 * seems like overkill.
1779 	 */
1780 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1781 		UDP6_INC_STATS(sock_net(sk),
1782 			       UDP_MIB_SNDBUFERRORS, is_udplite);
1783 	}
1784 	return err;
1785 
1786 do_confirm:
1787 	if (msg->msg_flags & MSG_PROBE)
1788 		dst_confirm_neigh(dst, &fl6->daddr);
1789 	if (!(msg->msg_flags&MSG_PROBE) || len)
1790 		goto back_from_confirm;
1791 	err = 0;
1792 	goto out;
1793 }
1794 EXPORT_SYMBOL(udpv6_sendmsg);
1795 
1796 static void udpv6_splice_eof(struct socket *sock)
1797 {
1798 	struct sock *sk = sock->sk;
1799 	struct udp_sock *up = udp_sk(sk);
1800 
1801 	if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1802 		return;
1803 
1804 	lock_sock(sk);
1805 	if (up->pending && !udp_test_bit(CORK, sk))
1806 		udp_v6_push_pending_frames(sk);
1807 	release_sock(sk);
1808 }
1809 
1810 void udpv6_destroy_sock(struct sock *sk)
1811 {
1812 	struct udp_sock *up = udp_sk(sk);
1813 	lock_sock(sk);
1814 
1815 	/* protects from races with udp_abort() */
1816 	sock_set_flag(sk, SOCK_DEAD);
1817 	udp_v6_flush_pending_frames(sk);
1818 	release_sock(sk);
1819 
1820 	if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1821 		if (up->encap_type) {
1822 			void (*encap_destroy)(struct sock *sk);
1823 			encap_destroy = READ_ONCE(up->encap_destroy);
1824 			if (encap_destroy)
1825 				encap_destroy(sk);
1826 		}
1827 		if (udp_test_bit(ENCAP_ENABLED, sk)) {
1828 			static_branch_dec(&udpv6_encap_needed_key);
1829 			udp_encap_disable();
1830 			udp_tunnel_cleanup_gro(sk);
1831 		}
1832 	}
1833 }
1834 
1835 /*
1836  *	Socket option code for UDP
1837  */
1838 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
1839 		     unsigned int optlen)
1840 {
1841 	if (level == SOL_UDP  ||  level == SOL_UDPLITE || level == SOL_SOCKET)
1842 		return udp_lib_setsockopt(sk, level, optname,
1843 					  optval, optlen,
1844 					  udp_v6_push_pending_frames);
1845 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
1846 }
1847 
1848 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1849 		     char __user *optval, int __user *optlen)
1850 {
1851 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1852 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1853 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
1854 }
1855 
1856 
1857 /* ------------------------------------------------------------------------ */
1858 #ifdef CONFIG_PROC_FS
1859 int udp6_seq_show(struct seq_file *seq, void *v)
1860 {
1861 	if (v == SEQ_START_TOKEN) {
1862 		seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1863 	} else {
1864 		int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1865 		const struct inet_sock *inet = inet_sk((const struct sock *)v);
1866 		__u16 srcp = ntohs(inet->inet_sport);
1867 		__u16 destp = ntohs(inet->inet_dport);
1868 		__ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1869 					  udp_rqueue_get(v), bucket);
1870 	}
1871 	return 0;
1872 }
1873 
1874 const struct seq_operations udp6_seq_ops = {
1875 	.start		= udp_seq_start,
1876 	.next		= udp_seq_next,
1877 	.stop		= udp_seq_stop,
1878 	.show		= udp6_seq_show,
1879 };
1880 EXPORT_SYMBOL(udp6_seq_ops);
1881 
1882 static struct udp_seq_afinfo udp6_seq_afinfo = {
1883 	.family		= AF_INET6,
1884 	.udp_table	= NULL,
1885 };
1886 
1887 int __net_init udp6_proc_init(struct net *net)
1888 {
1889 	if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1890 			sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1891 		return -ENOMEM;
1892 	return 0;
1893 }
1894 
1895 void udp6_proc_exit(struct net *net)
1896 {
1897 	remove_proc_entry("udp6", net->proc_net);
1898 }
1899 #endif /* CONFIG_PROC_FS */
1900 
1901 /* ------------------------------------------------------------------------ */
1902 
1903 struct proto udpv6_prot = {
1904 	.name			= "UDPv6",
1905 	.owner			= THIS_MODULE,
1906 	.close			= udp_lib_close,
1907 	.pre_connect		= udpv6_pre_connect,
1908 	.connect		= udpv6_connect,
1909 	.disconnect		= udp_disconnect,
1910 	.ioctl			= udp_ioctl,
1911 	.init			= udpv6_init_sock,
1912 	.destroy		= udpv6_destroy_sock,
1913 	.setsockopt		= udpv6_setsockopt,
1914 	.getsockopt		= udpv6_getsockopt,
1915 	.sendmsg		= udpv6_sendmsg,
1916 	.recvmsg		= udpv6_recvmsg,
1917 	.splice_eof		= udpv6_splice_eof,
1918 	.release_cb		= ip6_datagram_release_cb,
1919 	.hash			= udp_lib_hash,
1920 	.unhash			= udp_lib_unhash,
1921 	.rehash			= udp_v6_rehash,
1922 	.get_port		= udp_v6_get_port,
1923 	.put_port		= udp_lib_unhash,
1924 #ifdef CONFIG_BPF_SYSCALL
1925 	.psock_update_sk_prot	= udp_bpf_update_proto,
1926 #endif
1927 
1928 	.memory_allocated	= &net_aligned_data.udp_memory_allocated,
1929 	.per_cpu_fw_alloc	= &udp_memory_per_cpu_fw_alloc,
1930 
1931 	.sysctl_mem		= sysctl_udp_mem,
1932 	.sysctl_wmem_offset     = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1933 	.sysctl_rmem_offset     = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1934 	.obj_size		= sizeof(struct udp6_sock),
1935 	.ipv6_pinfo_offset = offsetof(struct udp6_sock, inet6),
1936 	.h.udp_table		= NULL,
1937 	.diag_destroy		= udp_abort,
1938 };
1939 
1940 static struct inet_protosw udpv6_protosw = {
1941 	.type =      SOCK_DGRAM,
1942 	.protocol =  IPPROTO_UDP,
1943 	.prot =      &udpv6_prot,
1944 	.ops =       &inet6_dgram_ops,
1945 	.flags =     INET_PROTOSW_PERMANENT,
1946 };
1947 
1948 int __init udpv6_init(void)
1949 {
1950 	int ret;
1951 
1952 	net_hotdata.udpv6_protocol = (struct inet6_protocol) {
1953 		.handler     = udpv6_rcv,
1954 		.err_handler = udpv6_err,
1955 		.flags	     = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
1956 	};
1957 	ret = inet6_add_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1958 	if (ret)
1959 		goto out;
1960 
1961 	ret = inet6_register_protosw(&udpv6_protosw);
1962 	if (ret)
1963 		goto out_udpv6_protocol;
1964 out:
1965 	return ret;
1966 
1967 out_udpv6_protocol:
1968 	inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1969 	goto out;
1970 }
1971 
1972 void udpv6_exit(void)
1973 {
1974 	inet6_unregister_protosw(&udpv6_protosw);
1975 	inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1976 }
1977