xref: /linux/net/ipv6/udp.c (revision dfecb0c5af3b07ebfa84be63a7a21bfc9e29a872)
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/aligned_data.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/seg6.h>
47 #include <net/tcp_states.h>
48 #include <net/ip6_checksum.h>
49 #include <net/ip6_tunnel.h>
50 #include <net/udp_tunnel.h>
51 #include <net/xfrm.h>
52 #include <net/inet_hashtables.h>
53 #include <net/inet6_hashtables.h>
54 #include <net/busy_poll.h>
55 #include <net/sock_reuseport.h>
56 #include <net/gro.h>
57 
58 #include <linux/proc_fs.h>
59 #include <linux/seq_file.h>
60 #include <trace/events/skb.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 static 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 static 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 static 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 sk_buff *skb)
348 {
349 	struct udp_table *udptable = net->ipv4.udp_table;
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 		sk = inet6_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
375 						saddr, sport, daddr, hnum, dif,
376 						udp6_ehashfn);
377 		if (sk) {
378 			result = sk;
379 			goto done;
380 		}
381 	}
382 
383 	/* Got non-wildcard socket or error on first lookup */
384 	if (result)
385 		goto done;
386 
387 	/* Lookup wildcard sockets */
388 	hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
389 	hslot2 = udp_hashslot2(udptable, hash2);
390 
391 	result = udp6_lib_lookup2(net, saddr, sport,
392 				  &in6addr_any, hnum, dif, sdif,
393 				  hslot2, skb);
394 	if (!IS_ERR_OR_NULL(result))
395 		goto done;
396 
397 	/* Cover address change/lookup/rehash race: see __udp4_lib_lookup() */
398 	result = udp6_lib_lookup1(net, saddr, sport, daddr, hnum, dif, sdif,
399 				  udptable);
400 
401 done:
402 	if (IS_ERR(result))
403 		return NULL;
404 	return result;
405 }
406 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
407 
408 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
409 					  __be16 sport, __be16 dport)
410 {
411 	const struct ipv6hdr *iph = ipv6_hdr(skb);
412 
413 	return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
414 				 &iph->daddr, dport, inet6_iif(skb),
415 				 inet6_sdif(skb), skb);
416 }
417 
418 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
419 				 __be16 sport, __be16 dport)
420 {
421 	const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
422 	const struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + offset);
423 	int iif, sdif;
424 
425 	inet6_get_iif_sdif(skb, &iif, &sdif);
426 
427 	return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
428 				 &iph->daddr, dport, iif, sdif, NULL);
429 }
430 
431 /* Must be called under rcu_read_lock().
432  * Does increment socket refcount.
433  */
434 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
435 struct sock *udp6_lib_lookup(const struct net *net, const struct in6_addr *saddr, __be16 sport,
436 			     const struct in6_addr *daddr, __be16 dport, int dif)
437 {
438 	struct sock *sk;
439 
440 	sk =  __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, 0, NULL);
441 	if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
442 		sk = NULL;
443 	return sk;
444 }
445 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
446 #endif
447 
448 /* do not use the scratch area len for jumbogram: their length exceeds the
449  * scratch area space; note that the IP6CB flags is still in the first
450  * cacheline, so checking for jumbograms is cheap
451  */
452 static int udp6_skb_len(struct sk_buff *skb)
453 {
454 	return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
455 }
456 
457 /*
458  *	This should be easy, if there is something there we
459  *	return it, otherwise we block.
460  */
461 
462 INDIRECT_CALLABLE_SCOPE
463 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
464 		  int flags)
465 {
466 	int off, is_udp4, err, peeking = flags & MSG_PEEK;
467 	struct ipv6_pinfo *np = inet6_sk(sk);
468 	struct inet_sock *inet = inet_sk(sk);
469 	struct udp_mib __percpu *mib;
470 	bool checksum_valid = false;
471 	unsigned int ulen, copied;
472 	struct sk_buff *skb;
473 
474 	if (flags & MSG_ERRQUEUE)
475 		return ipv6_recv_error(sk, msg, len);
476 
477 	if (np->rxopt.bits.rxpmtu && READ_ONCE(np->rxpmtu))
478 		return ipv6_recv_rxpmtu(sk, msg, len);
479 
480 try_again:
481 	off = sk_peek_offset(sk, flags);
482 	skb = __skb_recv_udp(sk, flags, &off, &err);
483 	if (!skb)
484 		return err;
485 
486 	ulen = udp6_skb_len(skb);
487 	copied = len;
488 	if (copied > ulen - off)
489 		copied = ulen - off;
490 	else if (copied < ulen)
491 		msg->msg_flags |= MSG_TRUNC;
492 
493 	is_udp4 = (skb->protocol == htons(ETH_P_IP));
494 	mib = __UDPX_MIB(sk, is_udp4);
495 
496 	/* If checksum is needed at all, try to do it while copying the
497 	 * data.  If the data is truncated, do it before the copy.
498 	 */
499 	if (copied < ulen || peeking) {
500 		checksum_valid = udp_skb_csum_unnecessary(skb) ||
501 				!__udp_lib_checksum_complete(skb);
502 		if (!checksum_valid)
503 			goto csum_copy_err;
504 	}
505 
506 	if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
507 		if (udp_skb_is_linear(skb))
508 			err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
509 		else
510 			err = skb_copy_datagram_msg(skb, off, msg, copied);
511 	} else {
512 		err = skb_copy_and_csum_datagram_msg(skb, off, msg);
513 		if (err == -EINVAL)
514 			goto csum_copy_err;
515 	}
516 	if (unlikely(err)) {
517 		if (!peeking) {
518 			udp_drops_inc(sk);
519 			SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
520 		}
521 		kfree_skb(skb);
522 		return err;
523 	}
524 	if (!peeking)
525 		SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
526 
527 	sock_recv_cmsgs(msg, sk, skb);
528 
529 	/* Copy the address. */
530 	if (msg->msg_name) {
531 		DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
532 		sin6->sin6_family = AF_INET6;
533 		sin6->sin6_port = udp_hdr(skb)->source;
534 		sin6->sin6_flowinfo = 0;
535 
536 		if (is_udp4) {
537 			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
538 					       &sin6->sin6_addr);
539 			sin6->sin6_scope_id = 0;
540 		} else {
541 			sin6->sin6_addr = ipv6_hdr(skb)->saddr;
542 			sin6->sin6_scope_id =
543 				ipv6_iface_scope_id(&sin6->sin6_addr,
544 						    inet6_iif(skb));
545 		}
546 		msg->msg_namelen = sizeof(*sin6);
547 
548 		BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
549 						      (struct sockaddr *)sin6,
550 						      &msg->msg_namelen);
551 	}
552 
553 	if (udp_test_bit(GRO_ENABLED, sk))
554 		udp_cmsg_recv(msg, sk, skb);
555 
556 	if (np->rxopt.all)
557 		ip6_datagram_recv_common_ctl(sk, msg, skb);
558 
559 	if (is_udp4) {
560 		if (inet_cmsg_flags(inet))
561 			ip_cmsg_recv_offset(msg, sk, skb,
562 					    sizeof(struct udphdr), off);
563 	} else {
564 		if (np->rxopt.all)
565 			ip6_datagram_recv_specific_ctl(sk, msg, skb);
566 	}
567 
568 	err = copied;
569 	if (flags & MSG_TRUNC)
570 		err = ulen;
571 
572 	skb_consume_udp(sk, skb, peeking ? -err : err);
573 	return err;
574 
575 csum_copy_err:
576 	if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
577 				 udp_skb_destructor)) {
578 		SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS);
579 		SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
580 	}
581 	kfree_skb_reason(skb, SKB_DROP_REASON_UDP_CSUM);
582 
583 	/* starting over for a new packet, but check if we need to yield */
584 	cond_resched();
585 	msg->msg_flags &= ~MSG_TRUNC;
586 	goto try_again;
587 }
588 
589 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
590 void udpv6_encap_enable(void)
591 {
592 	static_branch_inc(&udpv6_encap_needed_key);
593 }
594 EXPORT_SYMBOL(udpv6_encap_enable);
595 
596 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
597  * through error handlers in encapsulations looking for a match.
598  */
599 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb,
600 				      struct inet6_skb_parm *opt,
601 				      u8 type, u8 code, int offset, __be32 info)
602 {
603 	int i;
604 
605 	for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
606 		int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt,
607 			       u8 type, u8 code, int offset, __be32 info);
608 		const struct ip6_tnl_encap_ops *encap;
609 
610 		encap = rcu_dereference(ip6tun_encaps[i]);
611 		if (!encap)
612 			continue;
613 		handler = encap->err_handler;
614 		if (handler && !handler(skb, opt, type, code, offset, info))
615 			return 0;
616 	}
617 
618 	return -ENOENT;
619 }
620 
621 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
622  * reversing source and destination port: this will match tunnels that force the
623  * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
624  * lwtunnels might actually break this assumption by being configured with
625  * different destination ports on endpoints, in this case we won't be able to
626  * trace ICMP messages back to them.
627  *
628  * If this doesn't match any socket, probe tunnels with arbitrary destination
629  * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
630  * we've sent packets to won't necessarily match the local destination port.
631  *
632  * Then ask the tunnel implementation to match the error against a valid
633  * association.
634  *
635  * Return an error if we can't find a match, the socket if we need further
636  * processing, zero otherwise.
637  */
638 static struct sock *__udp6_lib_err_encap(struct net *net,
639 					 const struct ipv6hdr *hdr, int offset,
640 					 struct udphdr *uh,
641 					 struct sock *sk,
642 					 struct sk_buff *skb,
643 					 struct inet6_skb_parm *opt,
644 					 u8 type, u8 code, __be32 info)
645 {
646 	int (*lookup)(struct sock *sk, struct sk_buff *skb);
647 	int network_offset, transport_offset;
648 	struct udp_sock *up;
649 
650 	network_offset = skb_network_offset(skb);
651 	transport_offset = skb_transport_offset(skb);
652 
653 	/* Network header needs to point to the outer IPv6 header inside ICMP */
654 	skb_reset_network_header(skb);
655 
656 	/* Transport header needs to point to the UDP header */
657 	skb_set_transport_header(skb, offset);
658 
659 	if (sk) {
660 		up = udp_sk(sk);
661 
662 		lookup = READ_ONCE(up->encap_err_lookup);
663 		if (lookup && lookup(sk, skb))
664 			sk = NULL;
665 
666 		goto out;
667 	}
668 
669 	sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
670 			       &hdr->saddr, uh->dest,
671 			       inet6_iif(skb), 0, skb);
672 	if (sk) {
673 		up = udp_sk(sk);
674 
675 		lookup = READ_ONCE(up->encap_err_lookup);
676 		if (!lookup || lookup(sk, skb))
677 			sk = NULL;
678 	}
679 
680 out:
681 	if (!sk) {
682 		sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
683 							offset, info));
684 	}
685 
686 	skb_set_transport_header(skb, transport_offset);
687 	skb_set_network_header(skb, network_offset);
688 
689 	return sk;
690 }
691 
692 static int udpv6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
693 		     u8 type, u8 code, int offset, __be32 info)
694 {
695 	const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
696 	struct udphdr *uh = (struct udphdr *)(skb->data + offset);
697 	const struct in6_addr *saddr, *daddr;
698 	struct net *net = dev_net(skb->dev);
699 	struct ipv6_pinfo *np;
700 	bool tunnel = false;
701 	struct sock *sk;
702 	int harderr;
703 	int err;
704 
705 	daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr;
706 	saddr = &hdr->saddr;
707 	sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
708 			       inet6_iif(skb), inet6_sdif(skb), NULL);
709 
710 	if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
711 		/* No socket for error: try tunnels before discarding */
712 		if (static_branch_unlikely(&udpv6_encap_needed_key)) {
713 			sk = __udp6_lib_err_encap(net, hdr, offset, uh, sk, skb,
714 						  opt, type, code, info);
715 			if (!sk)
716 				return 0;
717 		} else
718 			sk = ERR_PTR(-ENOENT);
719 
720 		if (IS_ERR(sk)) {
721 			__ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
722 					  ICMP6_MIB_INERRORS);
723 			return PTR_ERR(sk);
724 		}
725 
726 		tunnel = true;
727 	}
728 
729 	harderr = icmpv6_err_convert(type, code, &err);
730 	np = inet6_sk(sk);
731 
732 	if (type == ICMPV6_PKT_TOOBIG) {
733 		if (!ip6_sk_accept_pmtu(sk))
734 			goto out;
735 		ip6_sk_update_pmtu(skb, sk, info);
736 		if (READ_ONCE(np->pmtudisc) != IPV6_PMTUDISC_DONT)
737 			harderr = 1;
738 	}
739 	if (type == NDISC_REDIRECT) {
740 		if (tunnel) {
741 			ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
742 				     READ_ONCE(sk->sk_mark),
743 				     sk_uid(sk));
744 		} else {
745 			ip6_sk_redirect(skb, sk);
746 		}
747 		goto out;
748 	}
749 
750 	/* Tunnels don't have an application socket: don't pass errors back */
751 	if (tunnel) {
752 		if (udp_sk(sk)->encap_err_rcv)
753 			udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest,
754 						  ntohl(info), (u8 *)(uh+1));
755 		goto out;
756 	}
757 
758 	if (!inet6_test_bit(RECVERR6, sk)) {
759 		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
760 			goto out;
761 	} else {
762 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
763 	}
764 
765 	sk->sk_err = err;
766 	sk_error_report(sk);
767 out:
768 	return 0;
769 }
770 
771 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
772 {
773 	int rc;
774 
775 	if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
776 		sock_rps_save_rxhash(sk, skb);
777 		sk_mark_napi_id(sk, skb);
778 		sk_incoming_cpu_update(sk);
779 	} else {
780 		sk_mark_napi_id_once(sk, skb);
781 	}
782 
783 	rc = __udp_enqueue_schedule_skb(sk, skb);
784 	if (rc < 0) {
785 		enum skb_drop_reason drop_reason;
786 		struct net *net = sock_net(sk);
787 
788 		/* Note that an ENOMEM error is charged twice */
789 		if (rc == -ENOMEM) {
790 			UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS);
791 			drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
792 		} else {
793 			UDP6_INC_STATS(net, UDP_MIB_MEMERRORS);
794 			drop_reason = SKB_DROP_REASON_PROTO_MEM;
795 		}
796 		UDP6_INC_STATS(net, UDP_MIB_INERRORS);
797 		trace_udp_fail_queue_rcv_skb(rc, sk, skb);
798 		sk_skb_reason_drop(sk, skb, drop_reason);
799 		return -1;
800 	}
801 
802 	return 0;
803 }
804 
805 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
806 {
807 	enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
808 	struct udp_sock *up = udp_sk(sk);
809 	struct net *net = sock_net(sk);
810 
811 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
812 		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
813 		goto drop;
814 	}
815 	nf_reset_ct(skb);
816 
817 	if (static_branch_unlikely(&udpv6_encap_needed_key) &&
818 	    READ_ONCE(up->encap_type)) {
819 		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
820 
821 		/*
822 		 * This is an encapsulation socket so pass the skb to
823 		 * the socket's udp_encap_rcv() hook. Otherwise, just
824 		 * fall through and pass this up the UDP socket.
825 		 * up->encap_rcv() returns the following value:
826 		 * =0 if skb was successfully passed to the encap
827 		 *    handler or was discarded by it.
828 		 * >0 if skb should be passed on to UDP.
829 		 * <0 if skb should be resubmitted as proto -N
830 		 */
831 
832 		/* if we're overly short, let UDP handle it */
833 		encap_rcv = READ_ONCE(up->encap_rcv);
834 		if (encap_rcv) {
835 			int ret;
836 
837 			/* Verify checksum before giving to encap */
838 			if (udp_lib_checksum_complete(skb))
839 				goto csum_error;
840 
841 			ret = encap_rcv(sk, skb);
842 			if (ret <= 0) {
843 				__UDP6_INC_STATS(net, UDP_MIB_INDATAGRAMS);
844 				return -ret;
845 			}
846 		}
847 
848 		/* FALLTHROUGH -- it's a UDP Packet */
849 	}
850 
851 	prefetch(&sk->sk_rmem_alloc);
852 	if (rcu_access_pointer(sk->sk_filter) &&
853 	    udp_lib_checksum_complete(skb))
854 		goto csum_error;
855 
856 	if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr), &drop_reason))
857 		goto drop;
858 
859 	udp_csum_pull_header(skb);
860 
861 	skb_dst_drop(skb);
862 
863 	return __udpv6_queue_rcv_skb(sk, skb);
864 
865 csum_error:
866 	drop_reason = SKB_DROP_REASON_UDP_CSUM;
867 	__UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS);
868 drop:
869 	__UDP6_INC_STATS(net, UDP_MIB_INERRORS);
870 	udp_drops_inc(sk);
871 	sk_skb_reason_drop(sk, skb, drop_reason);
872 	return -1;
873 }
874 
875 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
876 {
877 	struct sk_buff *next, *segs;
878 	int ret;
879 
880 	if (likely(!udp_unexpected_gso(sk, skb)))
881 		return udpv6_queue_rcv_one_skb(sk, skb);
882 
883 	__skb_push(skb, -skb_mac_offset(skb));
884 	segs = udp_rcv_segment(sk, skb, false);
885 	skb_list_walk_safe(segs, skb, next) {
886 		__skb_pull(skb, skb_transport_offset(skb));
887 
888 		udp_post_segment_fix_csum(skb);
889 		ret = udpv6_queue_rcv_one_skb(sk, skb);
890 		if (ret > 0)
891 			ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
892 						 true);
893 	}
894 	return 0;
895 }
896 
897 static bool __udp_v6_is_mcast_sock(struct net *net, const struct sock *sk,
898 				   __be16 loc_port, const struct in6_addr *loc_addr,
899 				   __be16 rmt_port, const struct in6_addr *rmt_addr,
900 				   int dif, int sdif, unsigned short hnum)
901 {
902 	const struct inet_sock *inet = inet_sk(sk);
903 
904 	if (!net_eq(sock_net(sk), net))
905 		return false;
906 
907 	if (udp_sk(sk)->udp_port_hash != hnum ||
908 	    sk->sk_family != PF_INET6 ||
909 	    (inet->inet_dport && inet->inet_dport != rmt_port) ||
910 	    (!ipv6_addr_any(&sk->sk_v6_daddr) &&
911 		    !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
912 	    !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) ||
913 	    (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
914 		    !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
915 		return false;
916 	if (!inet6_mc_check(sk, loc_addr, rmt_addr))
917 		return false;
918 	return true;
919 }
920 
921 static void udp6_csum_zero_error(struct sk_buff *skb)
922 {
923 	/* RFC 2460 section 8.1 says that we SHOULD log
924 	 * this error. Well, it is reasonable.
925 	 */
926 	net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
927 			    &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
928 			    &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
929 }
930 
931 /*
932  * Note: called only from the BH handler context,
933  * so we don't need to lock the hashes.
934  */
935 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
936 				    const struct in6_addr *saddr,
937 				    const struct in6_addr *daddr)
938 {
939 	struct udp_table *udptable = net->ipv4.udp_table;
940 	const struct udphdr *uh = udp_hdr(skb);
941 	unsigned int hash2, hash2_any, offset;
942 	unsigned short hnum = ntohs(uh->dest);
943 	struct sock *sk, *first = NULL;
944 	int sdif = inet6_sdif(skb);
945 	int dif = inet6_iif(skb);
946 	struct hlist_node *node;
947 	struct udp_hslot *hslot;
948 	struct sk_buff *nskb;
949 	bool use_hash2;
950 
951 	hash2_any = 0;
952 	hash2 = 0;
953 	hslot = udp_hashslot(udptable, net, hnum);
954 	use_hash2 = hslot->count > 10;
955 	offset = offsetof(typeof(*sk), sk_node);
956 
957 	if (use_hash2) {
958 		hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
959 			    udptable->mask;
960 		hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
961 start_lookup:
962 		hslot = &udptable->hash2[hash2].hslot;
963 		offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
964 	}
965 
966 	sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
967 		if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
968 					    uh->source, saddr, dif, sdif,
969 					    hnum))
970 			continue;
971 		/* If zero checksum and no_check is not on for
972 		 * the socket then skip it.
973 		 */
974 		if (!uh->check && !udp_get_no_check6_rx(sk))
975 			continue;
976 		if (!first) {
977 			first = sk;
978 			continue;
979 		}
980 		nskb = skb_clone(skb, GFP_ATOMIC);
981 		if (unlikely(!nskb)) {
982 			udp_drops_inc(sk);
983 			__UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS);
984 			__UDP6_INC_STATS(net, UDP_MIB_INERRORS);
985 			continue;
986 		}
987 
988 		if (udpv6_queue_rcv_skb(sk, nskb) > 0)
989 			consume_skb(nskb);
990 	}
991 
992 	/* Also lookup *:port if we are using hash2 and haven't done so yet. */
993 	if (use_hash2 && hash2 != hash2_any) {
994 		hash2 = hash2_any;
995 		goto start_lookup;
996 	}
997 
998 	if (first) {
999 		if (udpv6_queue_rcv_skb(first, skb) > 0)
1000 			consume_skb(skb);
1001 	} else {
1002 		kfree_skb(skb);
1003 		__UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI);
1004 	}
1005 	return 0;
1006 }
1007 
1008 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1009 {
1010 	if (udp_sk_rx_dst_set(sk, dst))
1011 		sk->sk_rx_dst_cookie = rt6_get_cookie(dst_rt6_info(dst));
1012 }
1013 
1014 /* wrapper for udp_queue_rcv_skb taking care of csum conversion and
1015  * return code conversion for ip layer consumption
1016  */
1017 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
1018 				struct udphdr *uh)
1019 {
1020 	int ret;
1021 
1022 	if (inet_get_convert_csum(sk) && uh->check)
1023 		skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo);
1024 
1025 	ret = udpv6_queue_rcv_skb(sk, skb);
1026 
1027 	/* a return value > 0 means to resubmit the input */
1028 	if (ret > 0)
1029 		return ret;
1030 	return 0;
1031 }
1032 
1033 static int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh)
1034 {
1035 	int err;
1036 
1037 	/* To support RFC 6936 (allow zero checksum in UDP/IPV6 for tunnels)
1038 	 * we accept a checksum of zero here. When we find the socket
1039 	 * for the UDP packet we'll check if that socket allows zero checksum
1040 	 * for IPv6 (set by socket option).
1041 	 *
1042 	 * Note, we are only interested in != 0 or == 0, thus the
1043 	 * force to int.
1044 	 */
1045 	err = (__force int)skb_checksum_init_zero_check(skb, IPPROTO_UDP, uh->check,
1046 							ip6_compute_pseudo);
1047 	if (err)
1048 		return err;
1049 
1050 	if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
1051 		/* If SW calculated the value, we know it's bad */
1052 		if (skb->csum_complete_sw)
1053 			return 1;
1054 
1055 		/* HW says the value is bad. Let's validate that.
1056 		 * skb->csum is no longer the full packet checksum,
1057 		 * so don't treat is as such.
1058 		 */
1059 		skb_checksum_complete_unset(skb);
1060 	}
1061 
1062 	return 0;
1063 }
1064 
1065 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
1066 {
1067 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1068 	const struct in6_addr *saddr, *daddr;
1069 	struct net *net = dev_net(skb->dev);
1070 	struct sock *sk = NULL;
1071 	struct udphdr *uh;
1072 	bool refcounted;
1073 	u32 ulen = 0;
1074 
1075 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1076 		goto discard;
1077 
1078 	saddr = &ipv6_hdr(skb)->saddr;
1079 	daddr = &ipv6_hdr(skb)->daddr;
1080 	uh = udp_hdr(skb);
1081 
1082 	ulen = ntohs(uh->len);
1083 	if (ulen > skb->len)
1084 		goto short_packet;
1085 
1086 	/* Check for jumbo payload */
1087 	if (ulen == 0)
1088 		ulen = skb->len;
1089 
1090 	if (ulen < sizeof(*uh))
1091 		goto short_packet;
1092 
1093 	if (ulen < skb->len) {
1094 		if (pskb_trim_rcsum(skb, ulen))
1095 			goto short_packet;
1096 
1097 		saddr = &ipv6_hdr(skb)->saddr;
1098 		daddr = &ipv6_hdr(skb)->daddr;
1099 		uh = udp_hdr(skb);
1100 	}
1101 
1102 	if (udp6_csum_init(skb, uh))
1103 		goto csum_error;
1104 
1105 	/* Check if the socket is already available, e.g. due to early demux */
1106 	sk = inet6_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
1107 			      &refcounted, udp6_ehashfn);
1108 	if (IS_ERR(sk))
1109 		goto no_sk;
1110 
1111 	if (sk) {
1112 		struct dst_entry *dst = skb_dst(skb);
1113 		int ret;
1114 
1115 		if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
1116 			udp6_sk_rx_dst_set(sk, dst);
1117 
1118 		if (!uh->check && !udp_get_no_check6_rx(sk)) {
1119 			if (refcounted)
1120 				sock_put(sk);
1121 			goto report_csum_error;
1122 		}
1123 
1124 		ret = udp6_unicast_rcv_skb(sk, skb, uh);
1125 		if (refcounted)
1126 			sock_put(sk);
1127 		return ret;
1128 	}
1129 
1130 	/*
1131 	 *	Multicast receive code
1132 	 */
1133 	if (ipv6_addr_is_multicast(daddr))
1134 		return __udp6_lib_mcast_deliver(net, skb, saddr, daddr);
1135 
1136 	/* Unicast */
1137 	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest);
1138 	if (sk) {
1139 		if (!uh->check && !udp_get_no_check6_rx(sk))
1140 			goto report_csum_error;
1141 		return udp6_unicast_rcv_skb(sk, skb, uh);
1142 	}
1143 no_sk:
1144 	reason = SKB_DROP_REASON_NO_SOCKET;
1145 
1146 	if (!uh->check)
1147 		goto report_csum_error;
1148 
1149 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1150 		goto discard;
1151 	nf_reset_ct(skb);
1152 
1153 	if (udp_lib_checksum_complete(skb))
1154 		goto csum_error;
1155 
1156 	__UDP6_INC_STATS(net, UDP_MIB_NOPORTS);
1157 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
1158 
1159 	sk_skb_reason_drop(sk, skb, reason);
1160 	return 0;
1161 
1162 short_packet:
1163 	if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1164 		reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1165 	net_dbg_ratelimited("UDPv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
1166 			    saddr, ntohs(uh->source),
1167 			    ulen, skb->len,
1168 			    daddr, ntohs(uh->dest));
1169 	goto discard;
1170 
1171 report_csum_error:
1172 	udp6_csum_zero_error(skb);
1173 csum_error:
1174 	if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1175 		reason = SKB_DROP_REASON_UDP_CSUM;
1176 	__UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS);
1177 discard:
1178 	__UDP6_INC_STATS(net, UDP_MIB_INERRORS);
1179 	sk_skb_reason_drop(sk, skb, reason);
1180 	return 0;
1181 }
1182 
1183 
1184 static struct sock *__udp6_lib_demux_lookup(struct net *net,
1185 			__be16 loc_port, const struct in6_addr *loc_addr,
1186 			__be16 rmt_port, const struct in6_addr *rmt_addr,
1187 			int dif, int sdif)
1188 {
1189 	struct udp_table *udptable = net->ipv4.udp_table;
1190 	unsigned short hnum = ntohs(loc_port);
1191 	struct udp_hslot *hslot2;
1192 	unsigned int hash2;
1193 	__portpair ports;
1194 	struct sock *sk;
1195 
1196 	hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
1197 	hslot2 = udp_hashslot2(udptable, hash2);
1198 	ports = INET_COMBINED_PORTS(rmt_port, hnum);
1199 
1200 	udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1201 		if (sk->sk_state == TCP_ESTABLISHED &&
1202 		    inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif))
1203 			return sk;
1204 		/* Only check first socket in chain */
1205 		break;
1206 	}
1207 	return NULL;
1208 }
1209 
1210 void udp_v6_early_demux(struct sk_buff *skb)
1211 {
1212 	struct net *net = dev_net(skb->dev);
1213 	const struct udphdr *uh;
1214 	struct sock *sk;
1215 	struct dst_entry *dst;
1216 	int dif = skb->dev->ifindex;
1217 	int sdif = inet6_sdif(skb);
1218 
1219 	if (!pskb_may_pull(skb, skb_transport_offset(skb) +
1220 	    sizeof(struct udphdr)))
1221 		return;
1222 
1223 	uh = udp_hdr(skb);
1224 
1225 	if (skb->pkt_type == PACKET_HOST)
1226 		sk = __udp6_lib_demux_lookup(net, uh->dest,
1227 					     &ipv6_hdr(skb)->daddr,
1228 					     uh->source, &ipv6_hdr(skb)->saddr,
1229 					     dif, sdif);
1230 	else
1231 		return;
1232 
1233 	if (!sk)
1234 		return;
1235 
1236 	skb->sk = sk;
1237 	DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
1238 	skb->destructor = sock_pfree;
1239 	dst = rcu_dereference(sk->sk_rx_dst);
1240 
1241 	if (dst)
1242 		dst = dst_check(dst, sk->sk_rx_dst_cookie);
1243 	if (dst) {
1244 		/* set noref for now.
1245 		 * any place which wants to hold dst has to call
1246 		 * dst_hold_safe()
1247 		 */
1248 		skb_dst_set_noref(skb, dst);
1249 	}
1250 }
1251 
1252 /*
1253  * Throw away all pending data and cancel the corking. Socket is locked.
1254  */
1255 static void udp_v6_flush_pending_frames(struct sock *sk)
1256 {
1257 	struct udp_sock *up = udp_sk(sk);
1258 
1259 	if (up->pending == AF_INET)
1260 		udp_flush_pending_frames(sk);
1261 	else if (up->pending) {
1262 		up->len = 0;
1263 		WRITE_ONCE(up->pending, 0);
1264 		ip6_flush_pending_frames(sk);
1265 	}
1266 }
1267 
1268 static int udpv6_pre_connect(struct sock *sk, struct sockaddr_unsized *uaddr,
1269 			     int addr_len)
1270 {
1271 	if (addr_len < offsetofend(struct sockaddr, sa_family))
1272 		return -EINVAL;
1273 	/* The following checks are replicated from __ip6_datagram_connect()
1274 	 * and intended to prevent BPF program called below from accessing
1275 	 * bytes that are out of the bound specified by user in addr_len.
1276 	 */
1277 	if (uaddr->sa_family == AF_INET) {
1278 		if (ipv6_only_sock(sk))
1279 			return -EAFNOSUPPORT;
1280 		return udp_pre_connect(sk, uaddr, addr_len);
1281 	}
1282 
1283 	if (addr_len < SIN6_LEN_RFC2133)
1284 		return -EINVAL;
1285 
1286 	return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, &addr_len);
1287 }
1288 
1289 static int udpv6_connect(struct sock *sk, struct sockaddr_unsized *uaddr,
1290 			 int addr_len)
1291 {
1292 	int res;
1293 
1294 	lock_sock(sk);
1295 	res = __ip6_datagram_connect(sk, uaddr, addr_len);
1296 	if (!res)
1297 		udp6_hash4(sk);
1298 	release_sock(sk);
1299 	return res;
1300 }
1301 
1302 /**
1303  *	udp6_hwcsum_outgoing  -  handle outgoing HW checksumming
1304  *	@sk:	socket we are sending on
1305  *	@skb:	sk_buff containing the filled-in UDP header
1306  *		(checksum field must be zeroed out)
1307  *	@saddr: source address
1308  *	@daddr: destination address
1309  *	@len:	length of packet
1310  */
1311 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1312 				 const struct in6_addr *saddr,
1313 				 const struct in6_addr *daddr, int len)
1314 {
1315 	unsigned int offset;
1316 	struct udphdr *uh = udp_hdr(skb);
1317 	struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1318 	__wsum csum = 0;
1319 
1320 	if (!frags) {
1321 		/* Only one fragment on the socket.  */
1322 		skb->csum_start = skb_transport_header(skb) - skb->head;
1323 		skb->csum_offset = offsetof(struct udphdr, check);
1324 		uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1325 	} else {
1326 		/*
1327 		 * HW-checksum won't work as there are two or more
1328 		 * fragments on the socket so that all csums of sk_buffs
1329 		 * should be together
1330 		 */
1331 		offset = skb_transport_offset(skb);
1332 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1333 		csum = skb->csum;
1334 
1335 		skb->ip_summed = CHECKSUM_NONE;
1336 
1337 		do {
1338 			csum = csum_add(csum, frags->csum);
1339 		} while ((frags = frags->next));
1340 
1341 		uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1342 					    csum);
1343 		if (uh->check == 0)
1344 			uh->check = CSUM_MANGLED_0;
1345 	}
1346 }
1347 
1348 /*
1349  *	Sending
1350  */
1351 
1352 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1353 			   struct inet_cork *cork)
1354 {
1355 	struct sock *sk = skb->sk;
1356 	int offset, len, datalen;
1357 	struct udphdr *uh;
1358 	int err = 0;
1359 
1360 	offset = skb_transport_offset(skb);
1361 	len = skb->len - offset;
1362 	datalen = len - sizeof(*uh);
1363 
1364 	/*
1365 	 * Create a UDP header
1366 	 */
1367 	uh = udp_hdr(skb);
1368 	uh->source = fl6->fl6_sport;
1369 	uh->dest = fl6->fl6_dport;
1370 	uh->len = htons(len);
1371 	uh->check = 0;
1372 
1373 	if (cork->gso_size) {
1374 		const int hlen = skb_network_header_len(skb) +
1375 				 sizeof(struct udphdr);
1376 
1377 		if (hlen + min(datalen, cork->gso_size) > cork->fragsize) {
1378 			kfree_skb(skb);
1379 			return -EMSGSIZE;
1380 		}
1381 		if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1382 			kfree_skb(skb);
1383 			return -EINVAL;
1384 		}
1385 		if (udp_get_no_check6_tx(sk)) {
1386 			kfree_skb(skb);
1387 			return -EINVAL;
1388 		}
1389 		if (dst_xfrm(skb_dst(skb))) {
1390 			kfree_skb(skb);
1391 			return -EIO;
1392 		}
1393 
1394 		if (datalen > cork->gso_size) {
1395 			skb_shinfo(skb)->gso_size = cork->gso_size;
1396 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1397 			skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1398 								 cork->gso_size);
1399 
1400 			/* Don't checksum the payload, skb will get segmented */
1401 			goto csum_partial;
1402 		}
1403 	}
1404 
1405 	if (udp_get_no_check6_tx(sk)) {   /* UDP csum disabled */
1406 		skb->ip_summed = CHECKSUM_NONE;
1407 		goto send;
1408 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1409 csum_partial:
1410 		udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1411 		goto send;
1412 	}
1413 
1414 	/* add protocol-dependent pseudo-header */
1415 	uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1416 				    len, IPPROTO_UDP, udp_csum(skb));
1417 	if (uh->check == 0)
1418 		uh->check = CSUM_MANGLED_0;
1419 
1420 send:
1421 	err = ip6_send_skb(skb);
1422 	if (unlikely(err)) {
1423 		if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk)) {
1424 			UDP6_INC_STATS(sock_net(sk), UDP_MIB_SNDBUFERRORS);
1425 			err = 0;
1426 		}
1427 	} else {
1428 		UDP6_INC_STATS(sock_net(sk), UDP_MIB_OUTDATAGRAMS);
1429 	}
1430 	return err;
1431 }
1432 
1433 static int udp_v6_push_pending_frames(struct sock *sk)
1434 {
1435 	struct sk_buff *skb;
1436 	struct udp_sock  *up = udp_sk(sk);
1437 	int err = 0;
1438 
1439 	if (up->pending == AF_INET)
1440 		return udp_push_pending_frames(sk);
1441 
1442 	skb = ip6_finish_skb(sk);
1443 	if (!skb)
1444 		goto out;
1445 
1446 	err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6,
1447 			      &inet_sk(sk)->cork.base);
1448 out:
1449 	up->len = 0;
1450 	WRITE_ONCE(up->pending, 0);
1451 	return err;
1452 }
1453 
1454 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1455 {
1456 	int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1457 	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1458 	struct ipv6_txoptions *opt_to_free = NULL;
1459 	struct in6_addr *daddr, *final_p, final;
1460 	struct ip6_flowlabel *flowlabel = NULL;
1461 	struct inet_sock *inet = inet_sk(sk);
1462 	struct ipv6_pinfo *np = inet6_sk(sk);
1463 	struct ipv6_txoptions *opt = NULL;
1464 	struct udp_sock *up = udp_sk(sk);
1465 	struct ipv6_txoptions opt_space;
1466 	int addr_len = msg->msg_namelen;
1467 	struct inet_cork_full cork;
1468 	struct ipcm6_cookie ipc6;
1469 	bool connected = false;
1470 	struct dst_entry *dst;
1471 	struct flowi6 *fl6;
1472 	int ulen = len;
1473 	int err;
1474 
1475 	fl6 = &cork.fl.u.ip6;
1476 	ipcm6_init_sk(&ipc6, sk);
1477 	ipc6.gso_size = READ_ONCE(up->gso_size);
1478 
1479 	/* destination address check */
1480 	if (sin6) {
1481 		if (addr_len < offsetof(struct sockaddr, sa_data))
1482 			return -EINVAL;
1483 
1484 		switch (sin6->sin6_family) {
1485 		case AF_INET6:
1486 			if (addr_len < SIN6_LEN_RFC2133)
1487 				return -EINVAL;
1488 			daddr = &sin6->sin6_addr;
1489 			if (ipv6_addr_any(daddr) &&
1490 			    ipv6_addr_v4mapped(&np->saddr))
1491 				ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1492 						       daddr);
1493 			break;
1494 		case AF_INET:
1495 			goto do_udp_sendmsg;
1496 		case AF_UNSPEC:
1497 			msg->msg_name = sin6 = NULL;
1498 			msg->msg_namelen = addr_len = 0;
1499 			daddr = NULL;
1500 			break;
1501 		default:
1502 			return -EINVAL;
1503 		}
1504 	} else if (!READ_ONCE(up->pending)) {
1505 		if (sk->sk_state != TCP_ESTABLISHED)
1506 			return -EDESTADDRREQ;
1507 		daddr = &sk->sk_v6_daddr;
1508 	} else
1509 		daddr = NULL;
1510 
1511 	if (daddr) {
1512 		if (ipv6_addr_v4mapped(daddr)) {
1513 			struct sockaddr_in sin;
1514 			sin.sin_family = AF_INET;
1515 			sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1516 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
1517 			msg->msg_name = &sin;
1518 			msg->msg_namelen = sizeof(sin);
1519 do_udp_sendmsg:
1520 			err = ipv6_only_sock(sk) ?
1521 				-ENETUNREACH : udp_sendmsg(sk, msg, len);
1522 			msg->msg_name = sin6;
1523 			msg->msg_namelen = addr_len;
1524 			return err;
1525 		}
1526 	}
1527 
1528 	/* Rough check on arithmetic overflow,
1529 	   better check is made in ip6_append_data().
1530 	   */
1531 	if (len > INT_MAX - sizeof(struct udphdr))
1532 		return -EMSGSIZE;
1533 
1534 	if (READ_ONCE(up->pending)) {
1535 		if (READ_ONCE(up->pending) == AF_INET)
1536 			return udp_sendmsg(sk, msg, len);
1537 		/*
1538 		 * There are pending frames.
1539 		 * The socket lock must be held while it's corked.
1540 		 */
1541 		lock_sock(sk);
1542 		if (likely(up->pending)) {
1543 			if (unlikely(up->pending != AF_INET6)) {
1544 				release_sock(sk);
1545 				return -EAFNOSUPPORT;
1546 			}
1547 			dst = NULL;
1548 			goto do_append_data;
1549 		}
1550 		release_sock(sk);
1551 	}
1552 	ulen += sizeof(struct udphdr);
1553 
1554 	memset(fl6, 0, sizeof(*fl6));
1555 
1556 	if (sin6) {
1557 		if (sin6->sin6_port == 0)
1558 			return -EINVAL;
1559 
1560 		fl6->fl6_dport = sin6->sin6_port;
1561 		daddr = &sin6->sin6_addr;
1562 
1563 		if (inet6_test_bit(SNDFLOW, sk)) {
1564 			fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1565 			if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) {
1566 				flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1567 				if (IS_ERR(flowlabel))
1568 					return -EINVAL;
1569 			}
1570 		}
1571 
1572 		/*
1573 		 * Otherwise it will be difficult to maintain
1574 		 * sk->sk_dst_cache.
1575 		 */
1576 		if (sk->sk_state == TCP_ESTABLISHED &&
1577 		    ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1578 			daddr = &sk->sk_v6_daddr;
1579 
1580 		if (addr_len >= sizeof(struct sockaddr_in6) &&
1581 		    sin6->sin6_scope_id &&
1582 		    __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1583 			fl6->flowi6_oif = sin6->sin6_scope_id;
1584 	} else {
1585 		if (sk->sk_state != TCP_ESTABLISHED)
1586 			return -EDESTADDRREQ;
1587 
1588 		fl6->fl6_dport = inet->inet_dport;
1589 		daddr = &sk->sk_v6_daddr;
1590 		fl6->flowlabel = np->flow_label;
1591 		connected = true;
1592 	}
1593 
1594 	if (!fl6->flowi6_oif)
1595 		fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if);
1596 
1597 	if (!fl6->flowi6_oif)
1598 		fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1599 
1600 	fl6->flowi6_uid = sk_uid(sk);
1601 
1602 	if (msg->msg_controllen) {
1603 		opt = &opt_space;
1604 		memset(opt, 0, sizeof(struct ipv6_txoptions));
1605 		opt->tot_len = sizeof(*opt);
1606 		ipc6.opt = opt;
1607 
1608 		err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1609 		if (err > 0) {
1610 			err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6,
1611 						    &ipc6);
1612 			connected = false;
1613 		}
1614 		if (err < 0) {
1615 			fl6_sock_release(flowlabel);
1616 			return err;
1617 		}
1618 		if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1619 			flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1620 			if (IS_ERR(flowlabel))
1621 				return -EINVAL;
1622 		}
1623 		if (!(opt->opt_nflen|opt->opt_flen))
1624 			opt = NULL;
1625 	}
1626 	if (!opt) {
1627 		opt = txopt_get(np);
1628 		opt_to_free = opt;
1629 	}
1630 	if (flowlabel)
1631 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
1632 	opt = ipv6_fixup_options(&opt_space, opt);
1633 	ipc6.opt = opt;
1634 
1635 	fl6->flowi6_proto = IPPROTO_UDP;
1636 	fl6->flowi6_mark = ipc6.sockc.mark;
1637 	fl6->daddr = *daddr;
1638 	if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr))
1639 		fl6->saddr = np->saddr;
1640 	fl6->fl6_sport = inet->inet_sport;
1641 
1642 	if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) {
1643 		err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1644 					   (struct sockaddr *)sin6,
1645 					   &addr_len,
1646 					   &fl6->saddr);
1647 		if (err)
1648 			goto out_no_dst;
1649 		if (sin6) {
1650 			if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1651 				/* BPF program rewrote IPv6-only by IPv4-mapped
1652 				 * IPv6. It's currently unsupported.
1653 				 */
1654 				err = -ENOTSUPP;
1655 				goto out_no_dst;
1656 			}
1657 			if (sin6->sin6_port == 0) {
1658 				/* BPF program set invalid port. Reject it. */
1659 				err = -EINVAL;
1660 				goto out_no_dst;
1661 			}
1662 			fl6->fl6_dport = sin6->sin6_port;
1663 			fl6->daddr = sin6->sin6_addr;
1664 		}
1665 	}
1666 
1667 	if (ipv6_addr_any(&fl6->daddr))
1668 		fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1669 
1670 	final_p = fl6_update_dst(fl6, opt, &final);
1671 	if (final_p)
1672 		connected = false;
1673 
1674 	if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) {
1675 		fl6->flowi6_oif = READ_ONCE(np->mcast_oif);
1676 		connected = false;
1677 	} else if (!fl6->flowi6_oif)
1678 		fl6->flowi6_oif = READ_ONCE(np->ucast_oif);
1679 
1680 	security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6));
1681 
1682 	fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel);
1683 
1684 	dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected);
1685 	if (IS_ERR(dst)) {
1686 		err = PTR_ERR(dst);
1687 		dst = NULL;
1688 		goto out;
1689 	}
1690 
1691 	if (ipc6.hlimit < 0)
1692 		ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst);
1693 
1694 	if (msg->msg_flags&MSG_CONFIRM)
1695 		goto do_confirm;
1696 back_from_confirm:
1697 
1698 	/* Lockless fast path for the non-corking case */
1699 	if (!corkreq) {
1700 		struct sk_buff *skb;
1701 
1702 		skb = ip6_make_skb(sk, ip_generic_getfrag, msg, ulen,
1703 				   sizeof(struct udphdr), &ipc6,
1704 				   dst_rt6_info(dst),
1705 				   msg->msg_flags, &cork);
1706 		err = PTR_ERR(skb);
1707 		if (!IS_ERR_OR_NULL(skb))
1708 			err = udp_v6_send_skb(skb, fl6, &cork.base);
1709 		/* ip6_make_skb steals dst reference */
1710 		goto out_no_dst;
1711 	}
1712 
1713 	lock_sock(sk);
1714 	if (unlikely(up->pending)) {
1715 		/* The socket is already corked while preparing it. */
1716 		/* ... which is an evident application bug. --ANK */
1717 		release_sock(sk);
1718 
1719 		net_dbg_ratelimited("udp cork app bug 2\n");
1720 		err = -EINVAL;
1721 		goto out;
1722 	}
1723 
1724 	WRITE_ONCE(up->pending, AF_INET6);
1725 
1726 do_append_data:
1727 	up->len += ulen;
1728 	err = ip6_append_data(sk, ip_generic_getfrag, msg, ulen,
1729 			      sizeof(struct udphdr), &ipc6, fl6,
1730 			      dst_rt6_info(dst),
1731 			      corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1732 	if (err)
1733 		udp_v6_flush_pending_frames(sk);
1734 	else if (!corkreq)
1735 		err = udp_v6_push_pending_frames(sk);
1736 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1737 		WRITE_ONCE(up->pending, 0);
1738 
1739 	if (err > 0)
1740 		err = inet6_test_bit(RECVERR6, sk) ? net_xmit_errno(err) : 0;
1741 	release_sock(sk);
1742 
1743 out:
1744 	dst_release(dst);
1745 out_no_dst:
1746 	fl6_sock_release(flowlabel);
1747 	txopt_put(opt_to_free);
1748 	if (!err)
1749 		return len;
1750 	/*
1751 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1752 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1753 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1754 	 * things).  We could add another new stat but at least for now that
1755 	 * seems like overkill.
1756 	 */
1757 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags))
1758 		UDP6_INC_STATS(sock_net(sk), UDP_MIB_SNDBUFERRORS);
1759 
1760 	return err;
1761 
1762 do_confirm:
1763 	if (msg->msg_flags & MSG_PROBE)
1764 		dst_confirm_neigh(dst, &fl6->daddr);
1765 	if (!(msg->msg_flags&MSG_PROBE) || len)
1766 		goto back_from_confirm;
1767 	err = 0;
1768 	goto out;
1769 }
1770 EXPORT_SYMBOL(udpv6_sendmsg);
1771 
1772 static void udpv6_splice_eof(struct socket *sock)
1773 {
1774 	struct sock *sk = sock->sk;
1775 	struct udp_sock *up = udp_sk(sk);
1776 
1777 	if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1778 		return;
1779 
1780 	lock_sock(sk);
1781 	if (up->pending && !udp_test_bit(CORK, sk))
1782 		udp_v6_push_pending_frames(sk);
1783 	release_sock(sk);
1784 }
1785 
1786 static void udpv6_destroy_sock(struct sock *sk)
1787 {
1788 	struct udp_sock *up = udp_sk(sk);
1789 	lock_sock(sk);
1790 
1791 	/* protects from races with udp_abort() */
1792 	sock_set_flag(sk, SOCK_DEAD);
1793 	udp_v6_flush_pending_frames(sk);
1794 	release_sock(sk);
1795 
1796 	if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1797 		if (up->encap_type) {
1798 			void (*encap_destroy)(struct sock *sk);
1799 			encap_destroy = READ_ONCE(up->encap_destroy);
1800 			if (encap_destroy)
1801 				encap_destroy(sk);
1802 		}
1803 		if (udp_test_bit(ENCAP_ENABLED, sk)) {
1804 			static_branch_dec(&udpv6_encap_needed_key);
1805 			udp_encap_disable();
1806 			udp_tunnel_cleanup_gro(sk);
1807 		}
1808 	}
1809 }
1810 
1811 /*
1812  *	Socket option code for UDP
1813  */
1814 static int udpv6_setsockopt(struct sock *sk, int level, int optname,
1815 			    sockptr_t optval, unsigned int optlen)
1816 {
1817 	if (level == SOL_UDP || level == SOL_SOCKET)
1818 		return udp_lib_setsockopt(sk, level, optname,
1819 					  optval, optlen,
1820 					  udp_v6_push_pending_frames);
1821 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
1822 }
1823 
1824 static int udpv6_getsockopt(struct sock *sk, int level, int optname,
1825 			    char __user *optval, int __user *optlen)
1826 {
1827 	if (level == SOL_UDP)
1828 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1829 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
1830 }
1831 
1832 
1833 /* ------------------------------------------------------------------------ */
1834 #ifdef CONFIG_PROC_FS
1835 static int udp6_seq_show(struct seq_file *seq, void *v)
1836 {
1837 	if (v == SEQ_START_TOKEN) {
1838 		seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1839 	} else {
1840 		int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1841 		const struct inet_sock *inet = inet_sk((const struct sock *)v);
1842 		__u16 srcp = ntohs(inet->inet_sport);
1843 		__u16 destp = ntohs(inet->inet_dport);
1844 		__ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1845 					  udp_rqueue_get(v), bucket);
1846 	}
1847 	return 0;
1848 }
1849 
1850 static const struct seq_operations udp6_seq_ops = {
1851 	.start		= udp_seq_start,
1852 	.next		= udp_seq_next,
1853 	.stop		= udp_seq_stop,
1854 	.show		= udp6_seq_show,
1855 };
1856 
1857 static struct udp_seq_afinfo udp6_seq_afinfo = {
1858 	.family		= AF_INET6,
1859 };
1860 
1861 int __net_init udp6_proc_init(struct net *net)
1862 {
1863 	if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1864 			sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1865 		return -ENOMEM;
1866 	return 0;
1867 }
1868 
1869 void udp6_proc_exit(struct net *net)
1870 {
1871 	remove_proc_entry("udp6", net->proc_net);
1872 }
1873 #endif /* CONFIG_PROC_FS */
1874 
1875 /* ------------------------------------------------------------------------ */
1876 
1877 struct proto udpv6_prot = {
1878 	.name			= "UDPv6",
1879 	.owner			= THIS_MODULE,
1880 	.close			= udp_lib_close,
1881 	.pre_connect		= udpv6_pre_connect,
1882 	.connect		= udpv6_connect,
1883 	.disconnect		= udp_disconnect,
1884 	.ioctl			= udp_ioctl,
1885 	.init			= udpv6_init_sock,
1886 	.destroy		= udpv6_destroy_sock,
1887 	.setsockopt		= udpv6_setsockopt,
1888 	.getsockopt		= udpv6_getsockopt,
1889 	.sendmsg		= udpv6_sendmsg,
1890 	.recvmsg		= udpv6_recvmsg,
1891 	.splice_eof		= udpv6_splice_eof,
1892 	.release_cb		= ip6_datagram_release_cb,
1893 	.hash			= udp_lib_hash,
1894 	.unhash			= udp_lib_unhash,
1895 	.rehash			= udp_v6_rehash,
1896 	.get_port		= udp_v6_get_port,
1897 	.put_port		= udp_lib_unhash,
1898 #ifdef CONFIG_BPF_SYSCALL
1899 	.psock_update_sk_prot	= udp_bpf_update_proto,
1900 #endif
1901 
1902 	.memory_allocated	= &net_aligned_data.udp_memory_allocated,
1903 	.per_cpu_fw_alloc	= &udp_memory_per_cpu_fw_alloc,
1904 
1905 	.sysctl_mem		= sysctl_udp_mem,
1906 	.sysctl_wmem_offset     = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1907 	.sysctl_rmem_offset     = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1908 	.obj_size		= sizeof(struct udp6_sock),
1909 	.ipv6_pinfo_offset = offsetof(struct udp6_sock, inet6),
1910 	.diag_destroy		= udp_abort,
1911 };
1912 
1913 static struct inet_protosw udpv6_protosw = {
1914 	.type =      SOCK_DGRAM,
1915 	.protocol =  IPPROTO_UDP,
1916 	.prot =      &udpv6_prot,
1917 	.ops =       &inet6_dgram_ops,
1918 	.flags =     INET_PROTOSW_PERMANENT,
1919 };
1920 
1921 int __init udpv6_init(void)
1922 {
1923 	int ret;
1924 
1925 	net_hotdata.udpv6_protocol = (struct inet6_protocol) {
1926 		.handler     = udpv6_rcv,
1927 		.err_handler = udpv6_err,
1928 		.flags	     = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
1929 	};
1930 	ret = inet6_add_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1931 	if (ret)
1932 		goto out;
1933 
1934 	ret = inet6_register_protosw(&udpv6_protosw);
1935 	if (ret)
1936 		goto out_udpv6_protocol;
1937 out:
1938 	return ret;
1939 
1940 out_udpv6_protocol:
1941 	inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1942 	goto out;
1943 }
1944 
1945 void udpv6_exit(void)
1946 {
1947 	inet6_unregister_protosw(&udpv6_protosw);
1948 	inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1949 }
1950