xref: /openbmc/linux/net/ipv6/udp.c (revision 95e9fd10)
1 /*
2  *	UDP over IPv6
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	Based on linux/ipv4/udp.c
9  *
10  *	Fixes:
11  *	Hideaki YOSHIFUJI	:	sin6_scope_id support
12  *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
13  *	Alexey Kuznetsov		allow both IPv4 and IPv6 sockets to bind
14  *					a single port at the same time.
15  *      Kazunori MIYAZAWA @USAGI:       change process style to use ip6_append_data
16  *      YOSHIFUJI Hideaki @USAGI:	convert /proc/net/udp6 to seq_file.
17  *
18  *	This program is free software; you can redistribute it and/or
19  *      modify it under the terms of the GNU General Public License
20  *      as published by the Free Software Foundation; either version
21  *      2 of the License, or (at your option) any later version.
22  */
23 
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/socket.h>
27 #include <linux/sockios.h>
28 #include <linux/net.h>
29 #include <linux/in6.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/ipv6.h>
33 #include <linux/icmpv6.h>
34 #include <linux/init.h>
35 #include <linux/module.h>
36 #include <linux/skbuff.h>
37 #include <linux/slab.h>
38 #include <asm/uaccess.h>
39 
40 #include <net/ndisc.h>
41 #include <net/protocol.h>
42 #include <net/transp_v6.h>
43 #include <net/ip6_route.h>
44 #include <net/raw.h>
45 #include <net/tcp_states.h>
46 #include <net/ip6_checksum.h>
47 #include <net/xfrm.h>
48 
49 #include <linux/proc_fs.h>
50 #include <linux/seq_file.h>
51 #include <trace/events/skb.h>
52 #include "udp_impl.h"
53 
54 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
55 {
56 	const struct in6_addr *sk_rcv_saddr6 = &inet6_sk(sk)->rcv_saddr;
57 	const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
58 	__be32 sk1_rcv_saddr = sk_rcv_saddr(sk);
59 	__be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
60 	int sk_ipv6only = ipv6_only_sock(sk);
61 	int sk2_ipv6only = inet_v6_ipv6only(sk2);
62 	int addr_type = ipv6_addr_type(sk_rcv_saddr6);
63 	int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
64 
65 	/* if both are mapped, treat as IPv4 */
66 	if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
67 		return (!sk2_ipv6only &&
68 			(!sk1_rcv_saddr || !sk2_rcv_saddr ||
69 			  sk1_rcv_saddr == sk2_rcv_saddr));
70 
71 	if (addr_type2 == IPV6_ADDR_ANY &&
72 	    !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
73 		return 1;
74 
75 	if (addr_type == IPV6_ADDR_ANY &&
76 	    !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
77 		return 1;
78 
79 	if (sk2_rcv_saddr6 &&
80 	    ipv6_addr_equal(sk_rcv_saddr6, sk2_rcv_saddr6))
81 		return 1;
82 
83 	return 0;
84 }
85 
86 static unsigned int udp6_portaddr_hash(struct net *net,
87 				       const struct in6_addr *addr6,
88 				       unsigned int port)
89 {
90 	unsigned int hash, mix = net_hash_mix(net);
91 
92 	if (ipv6_addr_any(addr6))
93 		hash = jhash_1word(0, mix);
94 	else if (ipv6_addr_v4mapped(addr6))
95 		hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
96 	else
97 		hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
98 
99 	return hash ^ port;
100 }
101 
102 
103 int udp_v6_get_port(struct sock *sk, unsigned short snum)
104 {
105 	unsigned int hash2_nulladdr =
106 		udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
107 	unsigned int hash2_partial =
108 		udp6_portaddr_hash(sock_net(sk), &inet6_sk(sk)->rcv_saddr, 0);
109 
110 	/* precompute partial secondary hash */
111 	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
112 	return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
113 }
114 
115 static void udp_v6_rehash(struct sock *sk)
116 {
117 	u16 new_hash = udp6_portaddr_hash(sock_net(sk),
118 					  &inet6_sk(sk)->rcv_saddr,
119 					  inet_sk(sk)->inet_num);
120 
121 	udp_lib_rehash(sk, new_hash);
122 }
123 
124 static inline int compute_score(struct sock *sk, struct net *net,
125 				unsigned short hnum,
126 				const struct in6_addr *saddr, __be16 sport,
127 				const struct in6_addr *daddr, __be16 dport,
128 				int dif)
129 {
130 	int score = -1;
131 
132 	if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
133 			sk->sk_family == PF_INET6) {
134 		struct ipv6_pinfo *np = inet6_sk(sk);
135 		struct inet_sock *inet = inet_sk(sk);
136 
137 		score = 0;
138 		if (inet->inet_dport) {
139 			if (inet->inet_dport != sport)
140 				return -1;
141 			score++;
142 		}
143 		if (!ipv6_addr_any(&np->rcv_saddr)) {
144 			if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
145 				return -1;
146 			score++;
147 		}
148 		if (!ipv6_addr_any(&np->daddr)) {
149 			if (!ipv6_addr_equal(&np->daddr, saddr))
150 				return -1;
151 			score++;
152 		}
153 		if (sk->sk_bound_dev_if) {
154 			if (sk->sk_bound_dev_if != dif)
155 				return -1;
156 			score++;
157 		}
158 	}
159 	return score;
160 }
161 
162 #define SCORE2_MAX (1 + 1 + 1)
163 static inline int compute_score2(struct sock *sk, struct net *net,
164 				const struct in6_addr *saddr, __be16 sport,
165 				const struct in6_addr *daddr, unsigned short hnum,
166 				int dif)
167 {
168 	int score = -1;
169 
170 	if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
171 			sk->sk_family == PF_INET6) {
172 		struct ipv6_pinfo *np = inet6_sk(sk);
173 		struct inet_sock *inet = inet_sk(sk);
174 
175 		if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
176 			return -1;
177 		score = 0;
178 		if (inet->inet_dport) {
179 			if (inet->inet_dport != sport)
180 				return -1;
181 			score++;
182 		}
183 		if (!ipv6_addr_any(&np->daddr)) {
184 			if (!ipv6_addr_equal(&np->daddr, saddr))
185 				return -1;
186 			score++;
187 		}
188 		if (sk->sk_bound_dev_if) {
189 			if (sk->sk_bound_dev_if != dif)
190 				return -1;
191 			score++;
192 		}
193 	}
194 	return score;
195 }
196 
197 
198 /* called with read_rcu_lock() */
199 static struct sock *udp6_lib_lookup2(struct net *net,
200 		const struct in6_addr *saddr, __be16 sport,
201 		const struct in6_addr *daddr, unsigned int hnum, int dif,
202 		struct udp_hslot *hslot2, unsigned int slot2)
203 {
204 	struct sock *sk, *result;
205 	struct hlist_nulls_node *node;
206 	int score, badness;
207 
208 begin:
209 	result = NULL;
210 	badness = -1;
211 	udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
212 		score = compute_score2(sk, net, saddr, sport,
213 				      daddr, hnum, dif);
214 		if (score > badness) {
215 			result = sk;
216 			badness = score;
217 			if (score == SCORE2_MAX)
218 				goto exact_match;
219 		}
220 	}
221 	/*
222 	 * if the nulls value we got at the end of this lookup is
223 	 * not the expected one, we must restart lookup.
224 	 * We probably met an item that was moved to another chain.
225 	 */
226 	if (get_nulls_value(node) != slot2)
227 		goto begin;
228 
229 	if (result) {
230 exact_match:
231 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
232 			result = NULL;
233 		else if (unlikely(compute_score2(result, net, saddr, sport,
234 				  daddr, hnum, dif) < badness)) {
235 			sock_put(result);
236 			goto begin;
237 		}
238 	}
239 	return result;
240 }
241 
242 struct sock *__udp6_lib_lookup(struct net *net,
243 				      const struct in6_addr *saddr, __be16 sport,
244 				      const struct in6_addr *daddr, __be16 dport,
245 				      int dif, struct udp_table *udptable)
246 {
247 	struct sock *sk, *result;
248 	struct hlist_nulls_node *node;
249 	unsigned short hnum = ntohs(dport);
250 	unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
251 	struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
252 	int score, badness;
253 
254 	rcu_read_lock();
255 	if (hslot->count > 10) {
256 		hash2 = udp6_portaddr_hash(net, daddr, hnum);
257 		slot2 = hash2 & udptable->mask;
258 		hslot2 = &udptable->hash2[slot2];
259 		if (hslot->count < hslot2->count)
260 			goto begin;
261 
262 		result = udp6_lib_lookup2(net, saddr, sport,
263 					  daddr, hnum, dif,
264 					  hslot2, slot2);
265 		if (!result) {
266 			hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
267 			slot2 = hash2 & udptable->mask;
268 			hslot2 = &udptable->hash2[slot2];
269 			if (hslot->count < hslot2->count)
270 				goto begin;
271 
272 			result = udp6_lib_lookup2(net, saddr, sport,
273 						  &in6addr_any, hnum, dif,
274 						  hslot2, slot2);
275 		}
276 		rcu_read_unlock();
277 		return result;
278 	}
279 begin:
280 	result = NULL;
281 	badness = -1;
282 	sk_nulls_for_each_rcu(sk, node, &hslot->head) {
283 		score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
284 		if (score > badness) {
285 			result = sk;
286 			badness = score;
287 		}
288 	}
289 	/*
290 	 * if the nulls value we got at the end of this lookup is
291 	 * not the expected one, we must restart lookup.
292 	 * We probably met an item that was moved to another chain.
293 	 */
294 	if (get_nulls_value(node) != slot)
295 		goto begin;
296 
297 	if (result) {
298 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
299 			result = NULL;
300 		else if (unlikely(compute_score(result, net, hnum, saddr, sport,
301 					daddr, dport, dif) < badness)) {
302 			sock_put(result);
303 			goto begin;
304 		}
305 	}
306 	rcu_read_unlock();
307 	return result;
308 }
309 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
310 
311 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
312 					  __be16 sport, __be16 dport,
313 					  struct udp_table *udptable)
314 {
315 	struct sock *sk;
316 	const struct ipv6hdr *iph = ipv6_hdr(skb);
317 
318 	if (unlikely(sk = skb_steal_sock(skb)))
319 		return sk;
320 	return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
321 				 &iph->daddr, dport, inet6_iif(skb),
322 				 udptable);
323 }
324 
325 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
326 			     const struct in6_addr *daddr, __be16 dport, int dif)
327 {
328 	return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
329 }
330 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
331 
332 
333 /*
334  * 	This should be easy, if there is something there we
335  * 	return it, otherwise we block.
336  */
337 
338 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
339 		  struct msghdr *msg, size_t len,
340 		  int noblock, int flags, int *addr_len)
341 {
342 	struct ipv6_pinfo *np = inet6_sk(sk);
343 	struct inet_sock *inet = inet_sk(sk);
344 	struct sk_buff *skb;
345 	unsigned int ulen, copied;
346 	int peeked, off = 0;
347 	int err;
348 	int is_udplite = IS_UDPLITE(sk);
349 	int is_udp4;
350 	bool slow;
351 
352 	if (addr_len)
353 		*addr_len = sizeof(struct sockaddr_in6);
354 
355 	if (flags & MSG_ERRQUEUE)
356 		return ipv6_recv_error(sk, msg, len);
357 
358 	if (np->rxpmtu && np->rxopt.bits.rxpmtu)
359 		return ipv6_recv_rxpmtu(sk, msg, len);
360 
361 try_again:
362 	skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
363 				  &peeked, &off, &err);
364 	if (!skb)
365 		goto out;
366 
367 	ulen = skb->len - sizeof(struct udphdr);
368 	copied = len;
369 	if (copied > ulen)
370 		copied = ulen;
371 	else if (copied < ulen)
372 		msg->msg_flags |= MSG_TRUNC;
373 
374 	is_udp4 = (skb->protocol == htons(ETH_P_IP));
375 
376 	/*
377 	 * If checksum is needed at all, try to do it while copying the
378 	 * data.  If the data is truncated, or if we only want a partial
379 	 * coverage checksum (UDP-Lite), do it before the copy.
380 	 */
381 
382 	if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
383 		if (udp_lib_checksum_complete(skb))
384 			goto csum_copy_err;
385 	}
386 
387 	if (skb_csum_unnecessary(skb))
388 		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
389 					      msg->msg_iov, copied);
390 	else {
391 		err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
392 		if (err == -EINVAL)
393 			goto csum_copy_err;
394 	}
395 	if (unlikely(err)) {
396 		trace_kfree_skb(skb, udpv6_recvmsg);
397 		goto out_free;
398 	}
399 	if (!peeked) {
400 		if (is_udp4)
401 			UDP_INC_STATS_USER(sock_net(sk),
402 					UDP_MIB_INDATAGRAMS, is_udplite);
403 		else
404 			UDP6_INC_STATS_USER(sock_net(sk),
405 					UDP_MIB_INDATAGRAMS, is_udplite);
406 	}
407 
408 	sock_recv_ts_and_drops(msg, sk, skb);
409 
410 	/* Copy the address. */
411 	if (msg->msg_name) {
412 		struct sockaddr_in6 *sin6;
413 
414 		sin6 = (struct sockaddr_in6 *) msg->msg_name;
415 		sin6->sin6_family = AF_INET6;
416 		sin6->sin6_port = udp_hdr(skb)->source;
417 		sin6->sin6_flowinfo = 0;
418 		sin6->sin6_scope_id = 0;
419 
420 		if (is_udp4)
421 			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
422 					       &sin6->sin6_addr);
423 		else {
424 			sin6->sin6_addr = ipv6_hdr(skb)->saddr;
425 			if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
426 				sin6->sin6_scope_id = IP6CB(skb)->iif;
427 		}
428 
429 	}
430 	if (is_udp4) {
431 		if (inet->cmsg_flags)
432 			ip_cmsg_recv(msg, skb);
433 	} else {
434 		if (np->rxopt.all)
435 			datagram_recv_ctl(sk, msg, skb);
436 	}
437 
438 	err = copied;
439 	if (flags & MSG_TRUNC)
440 		err = ulen;
441 
442 out_free:
443 	skb_free_datagram_locked(sk, skb);
444 out:
445 	return err;
446 
447 csum_copy_err:
448 	slow = lock_sock_fast(sk);
449 	if (!skb_kill_datagram(sk, skb, flags)) {
450 		if (is_udp4)
451 			UDP_INC_STATS_USER(sock_net(sk),
452 					UDP_MIB_INERRORS, is_udplite);
453 		else
454 			UDP6_INC_STATS_USER(sock_net(sk),
455 					UDP_MIB_INERRORS, is_udplite);
456 	}
457 	unlock_sock_fast(sk, slow);
458 
459 	if (noblock)
460 		return -EAGAIN;
461 
462 	/* starting over for a new packet */
463 	msg->msg_flags &= ~MSG_TRUNC;
464 	goto try_again;
465 }
466 
467 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
468 		    u8 type, u8 code, int offset, __be32 info,
469 		    struct udp_table *udptable)
470 {
471 	struct ipv6_pinfo *np;
472 	const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
473 	const struct in6_addr *saddr = &hdr->saddr;
474 	const struct in6_addr *daddr = &hdr->daddr;
475 	struct udphdr *uh = (struct udphdr*)(skb->data+offset);
476 	struct sock *sk;
477 	int err;
478 
479 	sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
480 			       saddr, uh->source, inet6_iif(skb), udptable);
481 	if (sk == NULL)
482 		return;
483 
484 	if (type == ICMPV6_PKT_TOOBIG)
485 		ip6_sk_update_pmtu(skb, sk, info);
486 	if (type == NDISC_REDIRECT)
487 		ip6_sk_redirect(skb, sk);
488 
489 	np = inet6_sk(sk);
490 
491 	if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
492 		goto out;
493 
494 	if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
495 		goto out;
496 
497 	if (np->recverr)
498 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
499 
500 	sk->sk_err = err;
501 	sk->sk_error_report(sk);
502 out:
503 	sock_put(sk);
504 }
505 
506 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
507 {
508 	int rc;
509 
510 	if (!ipv6_addr_any(&inet6_sk(sk)->daddr))
511 		sock_rps_save_rxhash(sk, skb);
512 
513 	rc = sock_queue_rcv_skb(sk, skb);
514 	if (rc < 0) {
515 		int is_udplite = IS_UDPLITE(sk);
516 
517 		/* Note that an ENOMEM error is charged twice */
518 		if (rc == -ENOMEM)
519 			UDP6_INC_STATS_BH(sock_net(sk),
520 					UDP_MIB_RCVBUFERRORS, is_udplite);
521 		UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
522 		kfree_skb(skb);
523 		return -1;
524 	}
525 	return 0;
526 }
527 
528 static __inline__ void udpv6_err(struct sk_buff *skb,
529 				 struct inet6_skb_parm *opt, u8 type,
530 				 u8 code, int offset, __be32 info     )
531 {
532 	__udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
533 }
534 
535 static struct static_key udpv6_encap_needed __read_mostly;
536 void udpv6_encap_enable(void)
537 {
538 	if (!static_key_enabled(&udpv6_encap_needed))
539 		static_key_slow_inc(&udpv6_encap_needed);
540 }
541 EXPORT_SYMBOL(udpv6_encap_enable);
542 
543 int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
544 {
545 	struct udp_sock *up = udp_sk(sk);
546 	int rc;
547 	int is_udplite = IS_UDPLITE(sk);
548 
549 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
550 		goto drop;
551 
552 	if (static_key_false(&udpv6_encap_needed) && up->encap_type) {
553 		int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
554 
555 		/*
556 		 * This is an encapsulation socket so pass the skb to
557 		 * the socket's udp_encap_rcv() hook. Otherwise, just
558 		 * fall through and pass this up the UDP socket.
559 		 * up->encap_rcv() returns the following value:
560 		 * =0 if skb was successfully passed to the encap
561 		 *    handler or was discarded by it.
562 		 * >0 if skb should be passed on to UDP.
563 		 * <0 if skb should be resubmitted as proto -N
564 		 */
565 
566 		/* if we're overly short, let UDP handle it */
567 		encap_rcv = ACCESS_ONCE(up->encap_rcv);
568 		if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
569 			int ret;
570 
571 			ret = encap_rcv(sk, skb);
572 			if (ret <= 0) {
573 				UDP_INC_STATS_BH(sock_net(sk),
574 						 UDP_MIB_INDATAGRAMS,
575 						 is_udplite);
576 				return -ret;
577 			}
578 		}
579 
580 		/* FALLTHROUGH -- it's a UDP Packet */
581 	}
582 
583 	/*
584 	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
585 	 */
586 	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
587 
588 		if (up->pcrlen == 0) {          /* full coverage was set  */
589 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
590 				" %d while full coverage %d requested\n",
591 				UDP_SKB_CB(skb)->cscov, skb->len);
592 			goto drop;
593 		}
594 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
595 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
596 						    "too small, need min %d\n",
597 				       UDP_SKB_CB(skb)->cscov, up->pcrlen);
598 			goto drop;
599 		}
600 	}
601 
602 	if (rcu_access_pointer(sk->sk_filter)) {
603 		if (udp_lib_checksum_complete(skb))
604 			goto drop;
605 	}
606 
607 	if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
608 		goto drop;
609 
610 	skb_dst_drop(skb);
611 
612 	bh_lock_sock(sk);
613 	rc = 0;
614 	if (!sock_owned_by_user(sk))
615 		rc = __udpv6_queue_rcv_skb(sk, skb);
616 	else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
617 		bh_unlock_sock(sk);
618 		goto drop;
619 	}
620 	bh_unlock_sock(sk);
621 
622 	return rc;
623 drop:
624 	UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
625 	atomic_inc(&sk->sk_drops);
626 	kfree_skb(skb);
627 	return -1;
628 }
629 
630 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
631 				      __be16 loc_port, const struct in6_addr *loc_addr,
632 				      __be16 rmt_port, const struct in6_addr *rmt_addr,
633 				      int dif)
634 {
635 	struct hlist_nulls_node *node;
636 	struct sock *s = sk;
637 	unsigned short num = ntohs(loc_port);
638 
639 	sk_nulls_for_each_from(s, node) {
640 		struct inet_sock *inet = inet_sk(s);
641 
642 		if (!net_eq(sock_net(s), net))
643 			continue;
644 
645 		if (udp_sk(s)->udp_port_hash == num &&
646 		    s->sk_family == PF_INET6) {
647 			struct ipv6_pinfo *np = inet6_sk(s);
648 			if (inet->inet_dport) {
649 				if (inet->inet_dport != rmt_port)
650 					continue;
651 			}
652 			if (!ipv6_addr_any(&np->daddr) &&
653 			    !ipv6_addr_equal(&np->daddr, rmt_addr))
654 				continue;
655 
656 			if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
657 				continue;
658 
659 			if (!ipv6_addr_any(&np->rcv_saddr)) {
660 				if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
661 					continue;
662 			}
663 			if (!inet6_mc_check(s, loc_addr, rmt_addr))
664 				continue;
665 			return s;
666 		}
667 	}
668 	return NULL;
669 }
670 
671 static void flush_stack(struct sock **stack, unsigned int count,
672 			struct sk_buff *skb, unsigned int final)
673 {
674 	struct sk_buff *skb1 = NULL;
675 	struct sock *sk;
676 	unsigned int i;
677 
678 	for (i = 0; i < count; i++) {
679 		sk = stack[i];
680 		if (likely(skb1 == NULL))
681 			skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
682 		if (!skb1) {
683 			atomic_inc(&sk->sk_drops);
684 			UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
685 					  IS_UDPLITE(sk));
686 			UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
687 					  IS_UDPLITE(sk));
688 		}
689 
690 		if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0)
691 			skb1 = NULL;
692 	}
693 	if (unlikely(skb1))
694 		kfree_skb(skb1);
695 }
696 /*
697  * Note: called only from the BH handler context,
698  * so we don't need to lock the hashes.
699  */
700 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
701 		const struct in6_addr *saddr, const struct in6_addr *daddr,
702 		struct udp_table *udptable)
703 {
704 	struct sock *sk, *stack[256 / sizeof(struct sock *)];
705 	const struct udphdr *uh = udp_hdr(skb);
706 	struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
707 	int dif;
708 	unsigned int i, count = 0;
709 
710 	spin_lock(&hslot->lock);
711 	sk = sk_nulls_head(&hslot->head);
712 	dif = inet6_iif(skb);
713 	sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
714 	while (sk) {
715 		stack[count++] = sk;
716 		sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
717 				       uh->source, saddr, dif);
718 		if (unlikely(count == ARRAY_SIZE(stack))) {
719 			if (!sk)
720 				break;
721 			flush_stack(stack, count, skb, ~0);
722 			count = 0;
723 		}
724 	}
725 	/*
726 	 * before releasing the lock, we must take reference on sockets
727 	 */
728 	for (i = 0; i < count; i++)
729 		sock_hold(stack[i]);
730 
731 	spin_unlock(&hslot->lock);
732 
733 	if (count) {
734 		flush_stack(stack, count, skb, count - 1);
735 
736 		for (i = 0; i < count; i++)
737 			sock_put(stack[i]);
738 	} else {
739 		kfree_skb(skb);
740 	}
741 	return 0;
742 }
743 
744 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
745 				 int proto)
746 {
747 	int err;
748 
749 	UDP_SKB_CB(skb)->partial_cov = 0;
750 	UDP_SKB_CB(skb)->cscov = skb->len;
751 
752 	if (proto == IPPROTO_UDPLITE) {
753 		err = udplite_checksum_init(skb, uh);
754 		if (err)
755 			return err;
756 	}
757 
758 	if (uh->check == 0) {
759 		/* RFC 2460 section 8.1 says that we SHOULD log
760 		   this error. Well, it is reasonable.
761 		 */
762 		LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
763 		return 1;
764 	}
765 	if (skb->ip_summed == CHECKSUM_COMPLETE &&
766 	    !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
767 			     skb->len, proto, skb->csum))
768 		skb->ip_summed = CHECKSUM_UNNECESSARY;
769 
770 	if (!skb_csum_unnecessary(skb))
771 		skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
772 							 &ipv6_hdr(skb)->daddr,
773 							 skb->len, proto, 0));
774 
775 	return 0;
776 }
777 
778 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
779 		   int proto)
780 {
781 	struct net *net = dev_net(skb->dev);
782 	struct sock *sk;
783 	struct udphdr *uh;
784 	const struct in6_addr *saddr, *daddr;
785 	u32 ulen = 0;
786 
787 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
788 		goto discard;
789 
790 	saddr = &ipv6_hdr(skb)->saddr;
791 	daddr = &ipv6_hdr(skb)->daddr;
792 	uh = udp_hdr(skb);
793 
794 	ulen = ntohs(uh->len);
795 	if (ulen > skb->len)
796 		goto short_packet;
797 
798 	if (proto == IPPROTO_UDP) {
799 		/* UDP validates ulen. */
800 
801 		/* Check for jumbo payload */
802 		if (ulen == 0)
803 			ulen = skb->len;
804 
805 		if (ulen < sizeof(*uh))
806 			goto short_packet;
807 
808 		if (ulen < skb->len) {
809 			if (pskb_trim_rcsum(skb, ulen))
810 				goto short_packet;
811 			saddr = &ipv6_hdr(skb)->saddr;
812 			daddr = &ipv6_hdr(skb)->daddr;
813 			uh = udp_hdr(skb);
814 		}
815 	}
816 
817 	if (udp6_csum_init(skb, uh, proto))
818 		goto discard;
819 
820 	/*
821 	 *	Multicast receive code
822 	 */
823 	if (ipv6_addr_is_multicast(daddr))
824 		return __udp6_lib_mcast_deliver(net, skb,
825 				saddr, daddr, udptable);
826 
827 	/* Unicast */
828 
829 	/*
830 	 * check socket cache ... must talk to Alan about his plans
831 	 * for sock caches... i'll skip this for now.
832 	 */
833 	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
834 	if (sk != NULL) {
835 		int ret = udpv6_queue_rcv_skb(sk, skb);
836 		sock_put(sk);
837 
838 		/* a return value > 0 means to resubmit the input, but
839 		 * it wants the return to be -protocol, or 0
840 		 */
841 		if (ret > 0)
842 			return -ret;
843 
844 		return 0;
845 	}
846 
847 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
848 		goto discard;
849 
850 	if (udp_lib_checksum_complete(skb))
851 		goto discard;
852 
853 	UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
854 	icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
855 
856 	kfree_skb(skb);
857 	return 0;
858 
859 short_packet:
860 	LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
861 		       proto == IPPROTO_UDPLITE ? "-Lite" : "",
862 		       saddr,
863 		       ntohs(uh->source),
864 		       ulen,
865 		       skb->len,
866 		       daddr,
867 		       ntohs(uh->dest));
868 
869 discard:
870 	UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
871 	kfree_skb(skb);
872 	return 0;
873 }
874 
875 static __inline__ int udpv6_rcv(struct sk_buff *skb)
876 {
877 	return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
878 }
879 
880 /*
881  * Throw away all pending data and cancel the corking. Socket is locked.
882  */
883 static void udp_v6_flush_pending_frames(struct sock *sk)
884 {
885 	struct udp_sock *up = udp_sk(sk);
886 
887 	if (up->pending == AF_INET)
888 		udp_flush_pending_frames(sk);
889 	else if (up->pending) {
890 		up->len = 0;
891 		up->pending = 0;
892 		ip6_flush_pending_frames(sk);
893 	}
894 }
895 
896 /**
897  * 	udp6_hwcsum_outgoing  -  handle outgoing HW checksumming
898  * 	@sk: 	socket we are sending on
899  * 	@skb: 	sk_buff containing the filled-in UDP header
900  * 	        (checksum field must be zeroed out)
901  */
902 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
903 				 const struct in6_addr *saddr,
904 				 const struct in6_addr *daddr, int len)
905 {
906 	unsigned int offset;
907 	struct udphdr *uh = udp_hdr(skb);
908 	__wsum csum = 0;
909 
910 	if (skb_queue_len(&sk->sk_write_queue) == 1) {
911 		/* Only one fragment on the socket.  */
912 		skb->csum_start = skb_transport_header(skb) - skb->head;
913 		skb->csum_offset = offsetof(struct udphdr, check);
914 		uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
915 	} else {
916 		/*
917 		 * HW-checksum won't work as there are two or more
918 		 * fragments on the socket so that all csums of sk_buffs
919 		 * should be together
920 		 */
921 		offset = skb_transport_offset(skb);
922 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
923 
924 		skb->ip_summed = CHECKSUM_NONE;
925 
926 		skb_queue_walk(&sk->sk_write_queue, skb) {
927 			csum = csum_add(csum, skb->csum);
928 		}
929 
930 		uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
931 					    csum);
932 		if (uh->check == 0)
933 			uh->check = CSUM_MANGLED_0;
934 	}
935 }
936 
937 /*
938  *	Sending
939  */
940 
941 static int udp_v6_push_pending_frames(struct sock *sk)
942 {
943 	struct sk_buff *skb;
944 	struct udphdr *uh;
945 	struct udp_sock  *up = udp_sk(sk);
946 	struct inet_sock *inet = inet_sk(sk);
947 	struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
948 	int err = 0;
949 	int is_udplite = IS_UDPLITE(sk);
950 	__wsum csum = 0;
951 
952 	/* Grab the skbuff where UDP header space exists. */
953 	if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
954 		goto out;
955 
956 	/*
957 	 * Create a UDP header
958 	 */
959 	uh = udp_hdr(skb);
960 	uh->source = fl6->fl6_sport;
961 	uh->dest = fl6->fl6_dport;
962 	uh->len = htons(up->len);
963 	uh->check = 0;
964 
965 	if (is_udplite)
966 		csum = udplite_csum_outgoing(sk, skb);
967 	else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
968 		udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr,
969 				     up->len);
970 		goto send;
971 	} else
972 		csum = udp_csum_outgoing(sk, skb);
973 
974 	/* add protocol-dependent pseudo-header */
975 	uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
976 				    up->len, fl6->flowi6_proto, csum);
977 	if (uh->check == 0)
978 		uh->check = CSUM_MANGLED_0;
979 
980 send:
981 	err = ip6_push_pending_frames(sk);
982 	if (err) {
983 		if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
984 			UDP6_INC_STATS_USER(sock_net(sk),
985 					    UDP_MIB_SNDBUFERRORS, is_udplite);
986 			err = 0;
987 		}
988 	} else
989 		UDP6_INC_STATS_USER(sock_net(sk),
990 				    UDP_MIB_OUTDATAGRAMS, is_udplite);
991 out:
992 	up->len = 0;
993 	up->pending = 0;
994 	return err;
995 }
996 
997 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
998 		  struct msghdr *msg, size_t len)
999 {
1000 	struct ipv6_txoptions opt_space;
1001 	struct udp_sock *up = udp_sk(sk);
1002 	struct inet_sock *inet = inet_sk(sk);
1003 	struct ipv6_pinfo *np = inet6_sk(sk);
1004 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
1005 	struct in6_addr *daddr, *final_p, final;
1006 	struct ipv6_txoptions *opt = NULL;
1007 	struct ip6_flowlabel *flowlabel = NULL;
1008 	struct flowi6 fl6;
1009 	struct dst_entry *dst;
1010 	int addr_len = msg->msg_namelen;
1011 	int ulen = len;
1012 	int hlimit = -1;
1013 	int tclass = -1;
1014 	int dontfrag = -1;
1015 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1016 	int err;
1017 	int connected = 0;
1018 	int is_udplite = IS_UDPLITE(sk);
1019 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1020 
1021 	/* destination address check */
1022 	if (sin6) {
1023 		if (addr_len < offsetof(struct sockaddr, sa_data))
1024 			return -EINVAL;
1025 
1026 		switch (sin6->sin6_family) {
1027 		case AF_INET6:
1028 			if (addr_len < SIN6_LEN_RFC2133)
1029 				return -EINVAL;
1030 			daddr = &sin6->sin6_addr;
1031 			break;
1032 		case AF_INET:
1033 			goto do_udp_sendmsg;
1034 		case AF_UNSPEC:
1035 			msg->msg_name = sin6 = NULL;
1036 			msg->msg_namelen = addr_len = 0;
1037 			daddr = NULL;
1038 			break;
1039 		default:
1040 			return -EINVAL;
1041 		}
1042 	} else if (!up->pending) {
1043 		if (sk->sk_state != TCP_ESTABLISHED)
1044 			return -EDESTADDRREQ;
1045 		daddr = &np->daddr;
1046 	} else
1047 		daddr = NULL;
1048 
1049 	if (daddr) {
1050 		if (ipv6_addr_v4mapped(daddr)) {
1051 			struct sockaddr_in sin;
1052 			sin.sin_family = AF_INET;
1053 			sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1054 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
1055 			msg->msg_name = &sin;
1056 			msg->msg_namelen = sizeof(sin);
1057 do_udp_sendmsg:
1058 			if (__ipv6_only_sock(sk))
1059 				return -ENETUNREACH;
1060 			return udp_sendmsg(iocb, sk, msg, len);
1061 		}
1062 	}
1063 
1064 	if (up->pending == AF_INET)
1065 		return udp_sendmsg(iocb, sk, msg, len);
1066 
1067 	/* Rough check on arithmetic overflow,
1068 	   better check is made in ip6_append_data().
1069 	   */
1070 	if (len > INT_MAX - sizeof(struct udphdr))
1071 		return -EMSGSIZE;
1072 
1073 	if (up->pending) {
1074 		/*
1075 		 * There are pending frames.
1076 		 * The socket lock must be held while it's corked.
1077 		 */
1078 		lock_sock(sk);
1079 		if (likely(up->pending)) {
1080 			if (unlikely(up->pending != AF_INET6)) {
1081 				release_sock(sk);
1082 				return -EAFNOSUPPORT;
1083 			}
1084 			dst = NULL;
1085 			goto do_append_data;
1086 		}
1087 		release_sock(sk);
1088 	}
1089 	ulen += sizeof(struct udphdr);
1090 
1091 	memset(&fl6, 0, sizeof(fl6));
1092 
1093 	if (sin6) {
1094 		if (sin6->sin6_port == 0)
1095 			return -EINVAL;
1096 
1097 		fl6.fl6_dport = sin6->sin6_port;
1098 		daddr = &sin6->sin6_addr;
1099 
1100 		if (np->sndflow) {
1101 			fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1102 			if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1103 				flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1104 				if (flowlabel == NULL)
1105 					return -EINVAL;
1106 				daddr = &flowlabel->dst;
1107 			}
1108 		}
1109 
1110 		/*
1111 		 * Otherwise it will be difficult to maintain
1112 		 * sk->sk_dst_cache.
1113 		 */
1114 		if (sk->sk_state == TCP_ESTABLISHED &&
1115 		    ipv6_addr_equal(daddr, &np->daddr))
1116 			daddr = &np->daddr;
1117 
1118 		if (addr_len >= sizeof(struct sockaddr_in6) &&
1119 		    sin6->sin6_scope_id &&
1120 		    ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
1121 			fl6.flowi6_oif = sin6->sin6_scope_id;
1122 	} else {
1123 		if (sk->sk_state != TCP_ESTABLISHED)
1124 			return -EDESTADDRREQ;
1125 
1126 		fl6.fl6_dport = inet->inet_dport;
1127 		daddr = &np->daddr;
1128 		fl6.flowlabel = np->flow_label;
1129 		connected = 1;
1130 	}
1131 
1132 	if (!fl6.flowi6_oif)
1133 		fl6.flowi6_oif = sk->sk_bound_dev_if;
1134 
1135 	if (!fl6.flowi6_oif)
1136 		fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1137 
1138 	fl6.flowi6_mark = sk->sk_mark;
1139 
1140 	if (msg->msg_controllen) {
1141 		opt = &opt_space;
1142 		memset(opt, 0, sizeof(struct ipv6_txoptions));
1143 		opt->tot_len = sizeof(*opt);
1144 
1145 		err = datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt,
1146 					&hlimit, &tclass, &dontfrag);
1147 		if (err < 0) {
1148 			fl6_sock_release(flowlabel);
1149 			return err;
1150 		}
1151 		if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1152 			flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1153 			if (flowlabel == NULL)
1154 				return -EINVAL;
1155 		}
1156 		if (!(opt->opt_nflen|opt->opt_flen))
1157 			opt = NULL;
1158 		connected = 0;
1159 	}
1160 	if (opt == NULL)
1161 		opt = np->opt;
1162 	if (flowlabel)
1163 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
1164 	opt = ipv6_fixup_options(&opt_space, opt);
1165 
1166 	fl6.flowi6_proto = sk->sk_protocol;
1167 	if (!ipv6_addr_any(daddr))
1168 		fl6.daddr = *daddr;
1169 	else
1170 		fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1171 	if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1172 		fl6.saddr = np->saddr;
1173 	fl6.fl6_sport = inet->inet_sport;
1174 
1175 	final_p = fl6_update_dst(&fl6, opt, &final);
1176 	if (final_p)
1177 		connected = 0;
1178 
1179 	if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1180 		fl6.flowi6_oif = np->mcast_oif;
1181 		connected = 0;
1182 	} else if (!fl6.flowi6_oif)
1183 		fl6.flowi6_oif = np->ucast_oif;
1184 
1185 	security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1186 
1187 	dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, true);
1188 	if (IS_ERR(dst)) {
1189 		err = PTR_ERR(dst);
1190 		dst = NULL;
1191 		goto out;
1192 	}
1193 
1194 	if (hlimit < 0) {
1195 		if (ipv6_addr_is_multicast(&fl6.daddr))
1196 			hlimit = np->mcast_hops;
1197 		else
1198 			hlimit = np->hop_limit;
1199 		if (hlimit < 0)
1200 			hlimit = ip6_dst_hoplimit(dst);
1201 	}
1202 
1203 	if (tclass < 0)
1204 		tclass = np->tclass;
1205 
1206 	if (dontfrag < 0)
1207 		dontfrag = np->dontfrag;
1208 
1209 	if (msg->msg_flags&MSG_CONFIRM)
1210 		goto do_confirm;
1211 back_from_confirm:
1212 
1213 	lock_sock(sk);
1214 	if (unlikely(up->pending)) {
1215 		/* The socket is already corked while preparing it. */
1216 		/* ... which is an evident application bug. --ANK */
1217 		release_sock(sk);
1218 
1219 		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1220 		err = -EINVAL;
1221 		goto out;
1222 	}
1223 
1224 	up->pending = AF_INET6;
1225 
1226 do_append_data:
1227 	up->len += ulen;
1228 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
1229 	err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1230 		sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1231 		(struct rt6_info*)dst,
1232 		corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1233 	if (err)
1234 		udp_v6_flush_pending_frames(sk);
1235 	else if (!corkreq)
1236 		err = udp_v6_push_pending_frames(sk);
1237 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1238 		up->pending = 0;
1239 
1240 	if (dst) {
1241 		if (connected) {
1242 			ip6_dst_store(sk, dst,
1243 				      ipv6_addr_equal(&fl6.daddr, &np->daddr) ?
1244 				      &np->daddr : NULL,
1245 #ifdef CONFIG_IPV6_SUBTREES
1246 				      ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1247 				      &np->saddr :
1248 #endif
1249 				      NULL);
1250 		} else {
1251 			dst_release(dst);
1252 		}
1253 		dst = NULL;
1254 	}
1255 
1256 	if (err > 0)
1257 		err = np->recverr ? net_xmit_errno(err) : 0;
1258 	release_sock(sk);
1259 out:
1260 	dst_release(dst);
1261 	fl6_sock_release(flowlabel);
1262 	if (!err)
1263 		return len;
1264 	/*
1265 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1266 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1267 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1268 	 * things).  We could add another new stat but at least for now that
1269 	 * seems like overkill.
1270 	 */
1271 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1272 		UDP6_INC_STATS_USER(sock_net(sk),
1273 				UDP_MIB_SNDBUFERRORS, is_udplite);
1274 	}
1275 	return err;
1276 
1277 do_confirm:
1278 	dst_confirm(dst);
1279 	if (!(msg->msg_flags&MSG_PROBE) || len)
1280 		goto back_from_confirm;
1281 	err = 0;
1282 	goto out;
1283 }
1284 
1285 void udpv6_destroy_sock(struct sock *sk)
1286 {
1287 	lock_sock(sk);
1288 	udp_v6_flush_pending_frames(sk);
1289 	release_sock(sk);
1290 
1291 	inet6_destroy_sock(sk);
1292 }
1293 
1294 /*
1295  *	Socket option code for UDP
1296  */
1297 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1298 		     char __user *optval, unsigned int optlen)
1299 {
1300 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1301 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1302 					  udp_v6_push_pending_frames);
1303 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
1304 }
1305 
1306 #ifdef CONFIG_COMPAT
1307 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1308 			    char __user *optval, unsigned int optlen)
1309 {
1310 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1311 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1312 					  udp_v6_push_pending_frames);
1313 	return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1314 }
1315 #endif
1316 
1317 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1318 		     char __user *optval, int __user *optlen)
1319 {
1320 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1321 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1322 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
1323 }
1324 
1325 #ifdef CONFIG_COMPAT
1326 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1327 			    char __user *optval, int __user *optlen)
1328 {
1329 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1330 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1331 	return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1332 }
1333 #endif
1334 
1335 static int udp6_ufo_send_check(struct sk_buff *skb)
1336 {
1337 	const struct ipv6hdr *ipv6h;
1338 	struct udphdr *uh;
1339 
1340 	if (!pskb_may_pull(skb, sizeof(*uh)))
1341 		return -EINVAL;
1342 
1343 	ipv6h = ipv6_hdr(skb);
1344 	uh = udp_hdr(skb);
1345 
1346 	uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
1347 				     IPPROTO_UDP, 0);
1348 	skb->csum_start = skb_transport_header(skb) - skb->head;
1349 	skb->csum_offset = offsetof(struct udphdr, check);
1350 	skb->ip_summed = CHECKSUM_PARTIAL;
1351 	return 0;
1352 }
1353 
1354 static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb,
1355 	netdev_features_t features)
1356 {
1357 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1358 	unsigned int mss;
1359 	unsigned int unfrag_ip6hlen, unfrag_len;
1360 	struct frag_hdr *fptr;
1361 	u8 *mac_start, *prevhdr;
1362 	u8 nexthdr;
1363 	u8 frag_hdr_sz = sizeof(struct frag_hdr);
1364 	int offset;
1365 	__wsum csum;
1366 
1367 	mss = skb_shinfo(skb)->gso_size;
1368 	if (unlikely(skb->len <= mss))
1369 		goto out;
1370 
1371 	if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1372 		/* Packet is from an untrusted source, reset gso_segs. */
1373 		int type = skb_shinfo(skb)->gso_type;
1374 
1375 		if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1376 			     !(type & (SKB_GSO_UDP))))
1377 			goto out;
1378 
1379 		skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1380 
1381 		segs = NULL;
1382 		goto out;
1383 	}
1384 
1385 	/* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1386 	 * do checksum of UDP packets sent as multiple IP fragments.
1387 	 */
1388 	offset = skb_checksum_start_offset(skb);
1389 	csum = skb_checksum(skb, offset, skb->len - offset, 0);
1390 	offset += skb->csum_offset;
1391 	*(__sum16 *)(skb->data + offset) = csum_fold(csum);
1392 	skb->ip_summed = CHECKSUM_NONE;
1393 
1394 	/* Check if there is enough headroom to insert fragment header. */
1395 	if ((skb_mac_header(skb) < skb->head + frag_hdr_sz) &&
1396 	    pskb_expand_head(skb, frag_hdr_sz, 0, GFP_ATOMIC))
1397 		goto out;
1398 
1399 	/* Find the unfragmentable header and shift it left by frag_hdr_sz
1400 	 * bytes to insert fragment header.
1401 	 */
1402 	unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
1403 	nexthdr = *prevhdr;
1404 	*prevhdr = NEXTHDR_FRAGMENT;
1405 	unfrag_len = skb_network_header(skb) - skb_mac_header(skb) +
1406 		     unfrag_ip6hlen;
1407 	mac_start = skb_mac_header(skb);
1408 	memmove(mac_start-frag_hdr_sz, mac_start, unfrag_len);
1409 
1410 	skb->mac_header -= frag_hdr_sz;
1411 	skb->network_header -= frag_hdr_sz;
1412 
1413 	fptr = (struct frag_hdr *)(skb_network_header(skb) + unfrag_ip6hlen);
1414 	fptr->nexthdr = nexthdr;
1415 	fptr->reserved = 0;
1416 	ipv6_select_ident(fptr, (struct rt6_info *)skb_dst(skb));
1417 
1418 	/* Fragment the skb. ipv6 header and the remaining fields of the
1419 	 * fragment header are updated in ipv6_gso_segment()
1420 	 */
1421 	segs = skb_segment(skb, features);
1422 
1423 out:
1424 	return segs;
1425 }
1426 
1427 static const struct inet6_protocol udpv6_protocol = {
1428 	.handler	=	udpv6_rcv,
1429 	.err_handler	=	udpv6_err,
1430 	.gso_send_check =	udp6_ufo_send_check,
1431 	.gso_segment	=	udp6_ufo_fragment,
1432 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1433 };
1434 
1435 /* ------------------------------------------------------------------------ */
1436 #ifdef CONFIG_PROC_FS
1437 
1438 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
1439 {
1440 	struct inet_sock *inet = inet_sk(sp);
1441 	struct ipv6_pinfo *np = inet6_sk(sp);
1442 	const struct in6_addr *dest, *src;
1443 	__u16 destp, srcp;
1444 
1445 	dest  = &np->daddr;
1446 	src   = &np->rcv_saddr;
1447 	destp = ntohs(inet->inet_dport);
1448 	srcp  = ntohs(inet->inet_sport);
1449 	seq_printf(seq,
1450 		   "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1451 		   "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n",
1452 		   bucket,
1453 		   src->s6_addr32[0], src->s6_addr32[1],
1454 		   src->s6_addr32[2], src->s6_addr32[3], srcp,
1455 		   dest->s6_addr32[0], dest->s6_addr32[1],
1456 		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
1457 		   sp->sk_state,
1458 		   sk_wmem_alloc_get(sp),
1459 		   sk_rmem_alloc_get(sp),
1460 		   0, 0L, 0,
1461 		   sock_i_uid(sp), 0,
1462 		   sock_i_ino(sp),
1463 		   atomic_read(&sp->sk_refcnt), sp,
1464 		   atomic_read(&sp->sk_drops));
1465 }
1466 
1467 int udp6_seq_show(struct seq_file *seq, void *v)
1468 {
1469 	if (v == SEQ_START_TOKEN)
1470 		seq_printf(seq,
1471 			   "  sl  "
1472 			   "local_address                         "
1473 			   "remote_address                        "
1474 			   "st tx_queue rx_queue tr tm->when retrnsmt"
1475 			   "   uid  timeout inode ref pointer drops\n");
1476 	else
1477 		udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1478 	return 0;
1479 }
1480 
1481 static const struct file_operations udp6_afinfo_seq_fops = {
1482 	.owner    = THIS_MODULE,
1483 	.open     = udp_seq_open,
1484 	.read     = seq_read,
1485 	.llseek   = seq_lseek,
1486 	.release  = seq_release_net
1487 };
1488 
1489 static struct udp_seq_afinfo udp6_seq_afinfo = {
1490 	.name		= "udp6",
1491 	.family		= AF_INET6,
1492 	.udp_table	= &udp_table,
1493 	.seq_fops	= &udp6_afinfo_seq_fops,
1494 	.seq_ops	= {
1495 		.show		= udp6_seq_show,
1496 	},
1497 };
1498 
1499 int __net_init udp6_proc_init(struct net *net)
1500 {
1501 	return udp_proc_register(net, &udp6_seq_afinfo);
1502 }
1503 
1504 void udp6_proc_exit(struct net *net) {
1505 	udp_proc_unregister(net, &udp6_seq_afinfo);
1506 }
1507 #endif /* CONFIG_PROC_FS */
1508 
1509 /* ------------------------------------------------------------------------ */
1510 
1511 struct proto udpv6_prot = {
1512 	.name		   = "UDPv6",
1513 	.owner		   = THIS_MODULE,
1514 	.close		   = udp_lib_close,
1515 	.connect	   = ip6_datagram_connect,
1516 	.disconnect	   = udp_disconnect,
1517 	.ioctl		   = udp_ioctl,
1518 	.destroy	   = udpv6_destroy_sock,
1519 	.setsockopt	   = udpv6_setsockopt,
1520 	.getsockopt	   = udpv6_getsockopt,
1521 	.sendmsg	   = udpv6_sendmsg,
1522 	.recvmsg	   = udpv6_recvmsg,
1523 	.backlog_rcv	   = __udpv6_queue_rcv_skb,
1524 	.hash		   = udp_lib_hash,
1525 	.unhash		   = udp_lib_unhash,
1526 	.rehash		   = udp_v6_rehash,
1527 	.get_port	   = udp_v6_get_port,
1528 	.memory_allocated  = &udp_memory_allocated,
1529 	.sysctl_mem	   = sysctl_udp_mem,
1530 	.sysctl_wmem	   = &sysctl_udp_wmem_min,
1531 	.sysctl_rmem	   = &sysctl_udp_rmem_min,
1532 	.obj_size	   = sizeof(struct udp6_sock),
1533 	.slab_flags	   = SLAB_DESTROY_BY_RCU,
1534 	.h.udp_table	   = &udp_table,
1535 #ifdef CONFIG_COMPAT
1536 	.compat_setsockopt = compat_udpv6_setsockopt,
1537 	.compat_getsockopt = compat_udpv6_getsockopt,
1538 #endif
1539 	.clear_sk	   = sk_prot_clear_portaddr_nulls,
1540 };
1541 
1542 static struct inet_protosw udpv6_protosw = {
1543 	.type =      SOCK_DGRAM,
1544 	.protocol =  IPPROTO_UDP,
1545 	.prot =      &udpv6_prot,
1546 	.ops =       &inet6_dgram_ops,
1547 	.no_check =  UDP_CSUM_DEFAULT,
1548 	.flags =     INET_PROTOSW_PERMANENT,
1549 };
1550 
1551 
1552 int __init udpv6_init(void)
1553 {
1554 	int ret;
1555 
1556 	ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1557 	if (ret)
1558 		goto out;
1559 
1560 	ret = inet6_register_protosw(&udpv6_protosw);
1561 	if (ret)
1562 		goto out_udpv6_protocol;
1563 out:
1564 	return ret;
1565 
1566 out_udpv6_protocol:
1567 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1568 	goto out;
1569 }
1570 
1571 void udpv6_exit(void)
1572 {
1573 	inet6_unregister_protosw(&udpv6_protosw);
1574 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1575 }
1576