xref: /openbmc/linux/net/ipv6/udp.c (revision 261a9af6)
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 "udp_impl.h"
52 
53 int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2)
54 {
55 	const struct in6_addr *sk_rcv_saddr6 = &inet6_sk(sk)->rcv_saddr;
56 	const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
57 	__be32 sk1_rcv_saddr = sk_rcv_saddr(sk);
58 	__be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
59 	int sk_ipv6only = ipv6_only_sock(sk);
60 	int sk2_ipv6only = inet_v6_ipv6only(sk2);
61 	int addr_type = ipv6_addr_type(sk_rcv_saddr6);
62 	int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
63 
64 	/* if both are mapped, treat as IPv4 */
65 	if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED)
66 		return (!sk2_ipv6only &&
67 			(!sk1_rcv_saddr || !sk2_rcv_saddr ||
68 			  sk1_rcv_saddr == sk2_rcv_saddr));
69 
70 	if (addr_type2 == IPV6_ADDR_ANY &&
71 	    !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
72 		return 1;
73 
74 	if (addr_type == IPV6_ADDR_ANY &&
75 	    !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
76 		return 1;
77 
78 	if (sk2_rcv_saddr6 &&
79 	    ipv6_addr_equal(sk_rcv_saddr6, sk2_rcv_saddr6))
80 		return 1;
81 
82 	return 0;
83 }
84 
85 static unsigned int udp6_portaddr_hash(struct net *net,
86 				       const struct in6_addr *addr6,
87 				       unsigned int port)
88 {
89 	unsigned int hash, mix = net_hash_mix(net);
90 
91 	if (ipv6_addr_any(addr6))
92 		hash = jhash_1word(0, mix);
93 	else if (ipv6_addr_v4mapped(addr6))
94 		hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix);
95 	else
96 		hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix);
97 
98 	return hash ^ port;
99 }
100 
101 
102 int udp_v6_get_port(struct sock *sk, unsigned short snum)
103 {
104 	unsigned int hash2_nulladdr =
105 		udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
106 	unsigned int hash2_partial =
107 		udp6_portaddr_hash(sock_net(sk), &inet6_sk(sk)->rcv_saddr, 0);
108 
109 	/* precompute partial secondary hash */
110 	udp_sk(sk)->udp_portaddr_hash = hash2_partial;
111 	return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr);
112 }
113 
114 static void udp_v6_rehash(struct sock *sk)
115 {
116 	u16 new_hash = udp6_portaddr_hash(sock_net(sk),
117 					  &inet6_sk(sk)->rcv_saddr,
118 					  inet_sk(sk)->inet_num);
119 
120 	udp_lib_rehash(sk, new_hash);
121 }
122 
123 static inline int compute_score(struct sock *sk, struct net *net,
124 				unsigned short hnum,
125 				const struct in6_addr *saddr, __be16 sport,
126 				const struct in6_addr *daddr, __be16 dport,
127 				int dif)
128 {
129 	int score = -1;
130 
131 	if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
132 			sk->sk_family == PF_INET6) {
133 		struct ipv6_pinfo *np = inet6_sk(sk);
134 		struct inet_sock *inet = inet_sk(sk);
135 
136 		score = 0;
137 		if (inet->inet_dport) {
138 			if (inet->inet_dport != sport)
139 				return -1;
140 			score++;
141 		}
142 		if (!ipv6_addr_any(&np->rcv_saddr)) {
143 			if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
144 				return -1;
145 			score++;
146 		}
147 		if (!ipv6_addr_any(&np->daddr)) {
148 			if (!ipv6_addr_equal(&np->daddr, saddr))
149 				return -1;
150 			score++;
151 		}
152 		if (sk->sk_bound_dev_if) {
153 			if (sk->sk_bound_dev_if != dif)
154 				return -1;
155 			score++;
156 		}
157 	}
158 	return score;
159 }
160 
161 #define SCORE2_MAX (1 + 1 + 1)
162 static inline int compute_score2(struct sock *sk, struct net *net,
163 				const struct in6_addr *saddr, __be16 sport,
164 				const struct in6_addr *daddr, unsigned short hnum,
165 				int dif)
166 {
167 	int score = -1;
168 
169 	if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
170 			sk->sk_family == PF_INET6) {
171 		struct ipv6_pinfo *np = inet6_sk(sk);
172 		struct inet_sock *inet = inet_sk(sk);
173 
174 		if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
175 			return -1;
176 		score = 0;
177 		if (inet->inet_dport) {
178 			if (inet->inet_dport != sport)
179 				return -1;
180 			score++;
181 		}
182 		if (!ipv6_addr_any(&np->daddr)) {
183 			if (!ipv6_addr_equal(&np->daddr, saddr))
184 				return -1;
185 			score++;
186 		}
187 		if (sk->sk_bound_dev_if) {
188 			if (sk->sk_bound_dev_if != dif)
189 				return -1;
190 			score++;
191 		}
192 	}
193 	return score;
194 }
195 
196 
197 /* called with read_rcu_lock() */
198 static struct sock *udp6_lib_lookup2(struct net *net,
199 		const struct in6_addr *saddr, __be16 sport,
200 		const struct in6_addr *daddr, unsigned int hnum, int dif,
201 		struct udp_hslot *hslot2, unsigned int slot2)
202 {
203 	struct sock *sk, *result;
204 	struct hlist_nulls_node *node;
205 	int score, badness;
206 
207 begin:
208 	result = NULL;
209 	badness = -1;
210 	udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
211 		score = compute_score2(sk, net, saddr, sport,
212 				      daddr, hnum, dif);
213 		if (score > badness) {
214 			result = sk;
215 			badness = score;
216 			if (score == SCORE2_MAX)
217 				goto exact_match;
218 		}
219 	}
220 	/*
221 	 * if the nulls value we got at the end of this lookup is
222 	 * not the expected one, we must restart lookup.
223 	 * We probably met an item that was moved to another chain.
224 	 */
225 	if (get_nulls_value(node) != slot2)
226 		goto begin;
227 
228 	if (result) {
229 exact_match:
230 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
231 			result = NULL;
232 		else if (unlikely(compute_score2(result, net, saddr, sport,
233 				  daddr, hnum, dif) < badness)) {
234 			sock_put(result);
235 			goto begin;
236 		}
237 	}
238 	return result;
239 }
240 
241 static struct sock *__udp6_lib_lookup(struct net *net,
242 				      const struct in6_addr *saddr, __be16 sport,
243 				      const struct in6_addr *daddr, __be16 dport,
244 				      int dif, struct udp_table *udptable)
245 {
246 	struct sock *sk, *result;
247 	struct hlist_nulls_node *node;
248 	unsigned short hnum = ntohs(dport);
249 	unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
250 	struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
251 	int score, badness;
252 
253 	rcu_read_lock();
254 	if (hslot->count > 10) {
255 		hash2 = udp6_portaddr_hash(net, daddr, hnum);
256 		slot2 = hash2 & udptable->mask;
257 		hslot2 = &udptable->hash2[slot2];
258 		if (hslot->count < hslot2->count)
259 			goto begin;
260 
261 		result = udp6_lib_lookup2(net, saddr, sport,
262 					  daddr, hnum, dif,
263 					  hslot2, slot2);
264 		if (!result) {
265 			hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum);
266 			slot2 = hash2 & udptable->mask;
267 			hslot2 = &udptable->hash2[slot2];
268 			if (hslot->count < hslot2->count)
269 				goto begin;
270 
271 			result = udp6_lib_lookup2(net, saddr, sport,
272 						  &in6addr_any, hnum, dif,
273 						  hslot2, slot2);
274 		}
275 		rcu_read_unlock();
276 		return result;
277 	}
278 begin:
279 	result = NULL;
280 	badness = -1;
281 	sk_nulls_for_each_rcu(sk, node, &hslot->head) {
282 		score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
283 		if (score > badness) {
284 			result = sk;
285 			badness = score;
286 		}
287 	}
288 	/*
289 	 * if the nulls value we got at the end of this lookup is
290 	 * not the expected one, we must restart lookup.
291 	 * We probably met an item that was moved to another chain.
292 	 */
293 	if (get_nulls_value(node) != slot)
294 		goto begin;
295 
296 	if (result) {
297 		if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
298 			result = NULL;
299 		else if (unlikely(compute_score(result, net, hnum, saddr, sport,
300 					daddr, dport, dif) < badness)) {
301 			sock_put(result);
302 			goto begin;
303 		}
304 	}
305 	rcu_read_unlock();
306 	return result;
307 }
308 
309 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
310 					  __be16 sport, __be16 dport,
311 					  struct udp_table *udptable)
312 {
313 	struct sock *sk;
314 	const struct ipv6hdr *iph = ipv6_hdr(skb);
315 
316 	if (unlikely(sk = skb_steal_sock(skb)))
317 		return sk;
318 	return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
319 				 &iph->daddr, dport, inet6_iif(skb),
320 				 udptable);
321 }
322 
323 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
324 			     const struct in6_addr *daddr, __be16 dport, int dif)
325 {
326 	return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
327 }
328 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
329 
330 
331 /*
332  * 	This should be easy, if there is something there we
333  * 	return it, otherwise we block.
334  */
335 
336 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
337 		  struct msghdr *msg, size_t len,
338 		  int noblock, int flags, int *addr_len)
339 {
340 	struct ipv6_pinfo *np = inet6_sk(sk);
341 	struct inet_sock *inet = inet_sk(sk);
342 	struct sk_buff *skb;
343 	unsigned int ulen;
344 	int peeked;
345 	int err;
346 	int is_udplite = IS_UDPLITE(sk);
347 	int is_udp4;
348 	bool slow;
349 
350 	if (addr_len)
351 		*addr_len=sizeof(struct sockaddr_in6);
352 
353 	if (flags & MSG_ERRQUEUE)
354 		return ipv6_recv_error(sk, msg, len);
355 
356 	if (np->rxpmtu && np->rxopt.bits.rxpmtu)
357 		return ipv6_recv_rxpmtu(sk, msg, len);
358 
359 try_again:
360 	skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
361 				  &peeked, &err);
362 	if (!skb)
363 		goto out;
364 
365 	ulen = skb->len - sizeof(struct udphdr);
366 	if (len > ulen)
367 		len = ulen;
368 	else if (len < ulen)
369 		msg->msg_flags |= MSG_TRUNC;
370 
371 	is_udp4 = (skb->protocol == htons(ETH_P_IP));
372 
373 	/*
374 	 * If checksum is needed at all, try to do it while copying the
375 	 * data.  If the data is truncated, or if we only want a partial
376 	 * coverage checksum (UDP-Lite), do it before the copy.
377 	 */
378 
379 	if (len < ulen || UDP_SKB_CB(skb)->partial_cov) {
380 		if (udp_lib_checksum_complete(skb))
381 			goto csum_copy_err;
382 	}
383 
384 	if (skb_csum_unnecessary(skb))
385 		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
386 					      msg->msg_iov,len);
387 	else {
388 		err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
389 		if (err == -EINVAL)
390 			goto csum_copy_err;
391 	}
392 	if (err)
393 		goto out_free;
394 
395 	if (!peeked) {
396 		if (is_udp4)
397 			UDP_INC_STATS_USER(sock_net(sk),
398 					UDP_MIB_INDATAGRAMS, is_udplite);
399 		else
400 			UDP6_INC_STATS_USER(sock_net(sk),
401 					UDP_MIB_INDATAGRAMS, is_udplite);
402 	}
403 
404 	sock_recv_ts_and_drops(msg, sk, skb);
405 
406 	/* Copy the address. */
407 	if (msg->msg_name) {
408 		struct sockaddr_in6 *sin6;
409 
410 		sin6 = (struct sockaddr_in6 *) msg->msg_name;
411 		sin6->sin6_family = AF_INET6;
412 		sin6->sin6_port = udp_hdr(skb)->source;
413 		sin6->sin6_flowinfo = 0;
414 		sin6->sin6_scope_id = 0;
415 
416 		if (is_udp4)
417 			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
418 					       &sin6->sin6_addr);
419 		else {
420 			ipv6_addr_copy(&sin6->sin6_addr,
421 				       &ipv6_hdr(skb)->saddr);
422 			if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
423 				sin6->sin6_scope_id = IP6CB(skb)->iif;
424 		}
425 
426 	}
427 	if (is_udp4) {
428 		if (inet->cmsg_flags)
429 			ip_cmsg_recv(msg, skb);
430 	} else {
431 		if (np->rxopt.all)
432 			datagram_recv_ctl(sk, msg, skb);
433 	}
434 
435 	err = len;
436 	if (flags & MSG_TRUNC)
437 		err = ulen;
438 
439 out_free:
440 	skb_free_datagram_locked(sk, skb);
441 out:
442 	return err;
443 
444 csum_copy_err:
445 	slow = lock_sock_fast(sk);
446 	if (!skb_kill_datagram(sk, skb, flags)) {
447 		if (is_udp4)
448 			UDP_INC_STATS_USER(sock_net(sk),
449 					UDP_MIB_INERRORS, is_udplite);
450 		else
451 			UDP6_INC_STATS_USER(sock_net(sk),
452 					UDP_MIB_INERRORS, is_udplite);
453 	}
454 	unlock_sock_fast(sk, slow);
455 
456 	if (flags & MSG_DONTWAIT)
457 		return -EAGAIN;
458 	goto try_again;
459 }
460 
461 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
462 		    u8 type, u8 code, int offset, __be32 info,
463 		    struct udp_table *udptable)
464 {
465 	struct ipv6_pinfo *np;
466 	const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
467 	const struct in6_addr *saddr = &hdr->saddr;
468 	const struct in6_addr *daddr = &hdr->daddr;
469 	struct udphdr *uh = (struct udphdr*)(skb->data+offset);
470 	struct sock *sk;
471 	int err;
472 
473 	sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
474 			       saddr, uh->source, inet6_iif(skb), udptable);
475 	if (sk == NULL)
476 		return;
477 
478 	np = inet6_sk(sk);
479 
480 	if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
481 		goto out;
482 
483 	if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
484 		goto out;
485 
486 	if (np->recverr)
487 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
488 
489 	sk->sk_err = err;
490 	sk->sk_error_report(sk);
491 out:
492 	sock_put(sk);
493 }
494 
495 static __inline__ void udpv6_err(struct sk_buff *skb,
496 				 struct inet6_skb_parm *opt, u8 type,
497 				 u8 code, int offset, __be32 info     )
498 {
499 	__udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
500 }
501 
502 int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
503 {
504 	struct udp_sock *up = udp_sk(sk);
505 	int rc;
506 	int is_udplite = IS_UDPLITE(sk);
507 
508 	if (!ipv6_addr_any(&inet6_sk(sk)->daddr))
509 		sock_rps_save_rxhash(sk, skb->rxhash);
510 
511 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
512 		goto drop;
513 
514 	/*
515 	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
516 	 */
517 	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
518 
519 		if (up->pcrlen == 0) {          /* full coverage was set  */
520 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
521 				" %d while full coverage %d requested\n",
522 				UDP_SKB_CB(skb)->cscov, skb->len);
523 			goto drop;
524 		}
525 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
526 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
527 						    "too small, need min %d\n",
528 				       UDP_SKB_CB(skb)->cscov, up->pcrlen);
529 			goto drop;
530 		}
531 	}
532 
533 	if (rcu_dereference_raw(sk->sk_filter)) {
534 		if (udp_lib_checksum_complete(skb))
535 			goto drop;
536 	}
537 
538 	if ((rc = ip_queue_rcv_skb(sk, skb)) < 0) {
539 		/* Note that an ENOMEM error is charged twice */
540 		if (rc == -ENOMEM)
541 			UDP6_INC_STATS_BH(sock_net(sk),
542 					UDP_MIB_RCVBUFERRORS, is_udplite);
543 		goto drop_no_sk_drops_inc;
544 	}
545 
546 	return 0;
547 drop:
548 	atomic_inc(&sk->sk_drops);
549 drop_no_sk_drops_inc:
550 	UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
551 	kfree_skb(skb);
552 	return -1;
553 }
554 
555 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
556 				      __be16 loc_port, const struct in6_addr *loc_addr,
557 				      __be16 rmt_port, const struct in6_addr *rmt_addr,
558 				      int dif)
559 {
560 	struct hlist_nulls_node *node;
561 	struct sock *s = sk;
562 	unsigned short num = ntohs(loc_port);
563 
564 	sk_nulls_for_each_from(s, node) {
565 		struct inet_sock *inet = inet_sk(s);
566 
567 		if (!net_eq(sock_net(s), net))
568 			continue;
569 
570 		if (udp_sk(s)->udp_port_hash == num &&
571 		    s->sk_family == PF_INET6) {
572 			struct ipv6_pinfo *np = inet6_sk(s);
573 			if (inet->inet_dport) {
574 				if (inet->inet_dport != rmt_port)
575 					continue;
576 			}
577 			if (!ipv6_addr_any(&np->daddr) &&
578 			    !ipv6_addr_equal(&np->daddr, rmt_addr))
579 				continue;
580 
581 			if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
582 				continue;
583 
584 			if (!ipv6_addr_any(&np->rcv_saddr)) {
585 				if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
586 					continue;
587 			}
588 			if (!inet6_mc_check(s, loc_addr, rmt_addr))
589 				continue;
590 			return s;
591 		}
592 	}
593 	return NULL;
594 }
595 
596 static void flush_stack(struct sock **stack, unsigned int count,
597 			struct sk_buff *skb, unsigned int final)
598 {
599 	unsigned int i;
600 	struct sock *sk;
601 	struct sk_buff *skb1;
602 
603 	for (i = 0; i < count; i++) {
604 		skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
605 
606 		sk = stack[i];
607 		if (skb1) {
608 			if (sk_rcvqueues_full(sk, skb1)) {
609 				kfree_skb(skb1);
610 				goto drop;
611 			}
612 			bh_lock_sock(sk);
613 			if (!sock_owned_by_user(sk))
614 				udpv6_queue_rcv_skb(sk, skb1);
615 			else if (sk_add_backlog(sk, skb1)) {
616 				kfree_skb(skb1);
617 				bh_unlock_sock(sk);
618 				goto drop;
619 			}
620 			bh_unlock_sock(sk);
621 			continue;
622 		}
623 drop:
624 		atomic_inc(&sk->sk_drops);
625 		UDP6_INC_STATS_BH(sock_net(sk),
626 				UDP_MIB_RCVBUFERRORS, IS_UDPLITE(sk));
627 		UDP6_INC_STATS_BH(sock_net(sk),
628 				UDP_MIB_INERRORS, IS_UDPLITE(sk));
629 	}
630 }
631 /*
632  * Note: called only from the BH handler context,
633  * so we don't need to lock the hashes.
634  */
635 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
636 		const struct in6_addr *saddr, const struct in6_addr *daddr,
637 		struct udp_table *udptable)
638 {
639 	struct sock *sk, *stack[256 / sizeof(struct sock *)];
640 	const struct udphdr *uh = udp_hdr(skb);
641 	struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
642 	int dif;
643 	unsigned int i, count = 0;
644 
645 	spin_lock(&hslot->lock);
646 	sk = sk_nulls_head(&hslot->head);
647 	dif = inet6_iif(skb);
648 	sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
649 	while (sk) {
650 		stack[count++] = sk;
651 		sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
652 				       uh->source, saddr, dif);
653 		if (unlikely(count == ARRAY_SIZE(stack))) {
654 			if (!sk)
655 				break;
656 			flush_stack(stack, count, skb, ~0);
657 			count = 0;
658 		}
659 	}
660 	/*
661 	 * before releasing the lock, we must take reference on sockets
662 	 */
663 	for (i = 0; i < count; i++)
664 		sock_hold(stack[i]);
665 
666 	spin_unlock(&hslot->lock);
667 
668 	if (count) {
669 		flush_stack(stack, count, skb, count - 1);
670 
671 		for (i = 0; i < count; i++)
672 			sock_put(stack[i]);
673 	} else {
674 		kfree_skb(skb);
675 	}
676 	return 0;
677 }
678 
679 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
680 				 int proto)
681 {
682 	int err;
683 
684 	UDP_SKB_CB(skb)->partial_cov = 0;
685 	UDP_SKB_CB(skb)->cscov = skb->len;
686 
687 	if (proto == IPPROTO_UDPLITE) {
688 		err = udplite_checksum_init(skb, uh);
689 		if (err)
690 			return err;
691 	}
692 
693 	if (uh->check == 0) {
694 		/* RFC 2460 section 8.1 says that we SHOULD log
695 		   this error. Well, it is reasonable.
696 		 */
697 		LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
698 		return 1;
699 	}
700 	if (skb->ip_summed == CHECKSUM_COMPLETE &&
701 	    !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
702 			     skb->len, proto, skb->csum))
703 		skb->ip_summed = CHECKSUM_UNNECESSARY;
704 
705 	if (!skb_csum_unnecessary(skb))
706 		skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
707 							 &ipv6_hdr(skb)->daddr,
708 							 skb->len, proto, 0));
709 
710 	return 0;
711 }
712 
713 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
714 		   int proto)
715 {
716 	struct net *net = dev_net(skb->dev);
717 	struct sock *sk;
718 	struct udphdr *uh;
719 	const struct in6_addr *saddr, *daddr;
720 	u32 ulen = 0;
721 
722 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
723 		goto discard;
724 
725 	saddr = &ipv6_hdr(skb)->saddr;
726 	daddr = &ipv6_hdr(skb)->daddr;
727 	uh = udp_hdr(skb);
728 
729 	ulen = ntohs(uh->len);
730 	if (ulen > skb->len)
731 		goto short_packet;
732 
733 	if (proto == IPPROTO_UDP) {
734 		/* UDP validates ulen. */
735 
736 		/* Check for jumbo payload */
737 		if (ulen == 0)
738 			ulen = skb->len;
739 
740 		if (ulen < sizeof(*uh))
741 			goto short_packet;
742 
743 		if (ulen < skb->len) {
744 			if (pskb_trim_rcsum(skb, ulen))
745 				goto short_packet;
746 			saddr = &ipv6_hdr(skb)->saddr;
747 			daddr = &ipv6_hdr(skb)->daddr;
748 			uh = udp_hdr(skb);
749 		}
750 	}
751 
752 	if (udp6_csum_init(skb, uh, proto))
753 		goto discard;
754 
755 	/*
756 	 *	Multicast receive code
757 	 */
758 	if (ipv6_addr_is_multicast(daddr))
759 		return __udp6_lib_mcast_deliver(net, skb,
760 				saddr, daddr, udptable);
761 
762 	/* Unicast */
763 
764 	/*
765 	 * check socket cache ... must talk to Alan about his plans
766 	 * for sock caches... i'll skip this for now.
767 	 */
768 	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
769 
770 	if (sk == NULL) {
771 		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
772 			goto discard;
773 
774 		if (udp_lib_checksum_complete(skb))
775 			goto discard;
776 		UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS,
777 				proto == IPPROTO_UDPLITE);
778 
779 		icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
780 
781 		kfree_skb(skb);
782 		return 0;
783 	}
784 
785 	/* deliver */
786 
787 	if (sk_rcvqueues_full(sk, skb)) {
788 		sock_put(sk);
789 		goto discard;
790 	}
791 	bh_lock_sock(sk);
792 	if (!sock_owned_by_user(sk))
793 		udpv6_queue_rcv_skb(sk, skb);
794 	else if (sk_add_backlog(sk, skb)) {
795 		atomic_inc(&sk->sk_drops);
796 		bh_unlock_sock(sk);
797 		sock_put(sk);
798 		goto discard;
799 	}
800 	bh_unlock_sock(sk);
801 	sock_put(sk);
802 	return 0;
803 
804 short_packet:
805 	LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
806 		       proto == IPPROTO_UDPLITE ? "-Lite" : "",
807 		       saddr,
808 		       ntohs(uh->source),
809 		       ulen,
810 		       skb->len,
811 		       daddr,
812 		       ntohs(uh->dest));
813 
814 discard:
815 	UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
816 	kfree_skb(skb);
817 	return 0;
818 }
819 
820 static __inline__ int udpv6_rcv(struct sk_buff *skb)
821 {
822 	return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
823 }
824 
825 /*
826  * Throw away all pending data and cancel the corking. Socket is locked.
827  */
828 static void udp_v6_flush_pending_frames(struct sock *sk)
829 {
830 	struct udp_sock *up = udp_sk(sk);
831 
832 	if (up->pending == AF_INET)
833 		udp_flush_pending_frames(sk);
834 	else if (up->pending) {
835 		up->len = 0;
836 		up->pending = 0;
837 		ip6_flush_pending_frames(sk);
838 	}
839 }
840 
841 /**
842  * 	udp6_hwcsum_outgoing  -  handle outgoing HW checksumming
843  * 	@sk: 	socket we are sending on
844  * 	@skb: 	sk_buff containing the filled-in UDP header
845  * 	        (checksum field must be zeroed out)
846  */
847 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
848 				 const struct in6_addr *saddr,
849 				 const struct in6_addr *daddr, int len)
850 {
851 	unsigned int offset;
852 	struct udphdr *uh = udp_hdr(skb);
853 	__wsum csum = 0;
854 
855 	if (skb_queue_len(&sk->sk_write_queue) == 1) {
856 		/* Only one fragment on the socket.  */
857 		skb->csum_start = skb_transport_header(skb) - skb->head;
858 		skb->csum_offset = offsetof(struct udphdr, check);
859 		uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
860 	} else {
861 		/*
862 		 * HW-checksum won't work as there are two or more
863 		 * fragments on the socket so that all csums of sk_buffs
864 		 * should be together
865 		 */
866 		offset = skb_transport_offset(skb);
867 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
868 
869 		skb->ip_summed = CHECKSUM_NONE;
870 
871 		skb_queue_walk(&sk->sk_write_queue, skb) {
872 			csum = csum_add(csum, skb->csum);
873 		}
874 
875 		uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
876 					    csum);
877 		if (uh->check == 0)
878 			uh->check = CSUM_MANGLED_0;
879 	}
880 }
881 
882 /*
883  *	Sending
884  */
885 
886 static int udp_v6_push_pending_frames(struct sock *sk)
887 {
888 	struct sk_buff *skb;
889 	struct udphdr *uh;
890 	struct udp_sock  *up = udp_sk(sk);
891 	struct inet_sock *inet = inet_sk(sk);
892 	struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
893 	int err = 0;
894 	int is_udplite = IS_UDPLITE(sk);
895 	__wsum csum = 0;
896 
897 	/* Grab the skbuff where UDP header space exists. */
898 	if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
899 		goto out;
900 
901 	/*
902 	 * Create a UDP header
903 	 */
904 	uh = udp_hdr(skb);
905 	uh->source = fl6->fl6_sport;
906 	uh->dest = fl6->fl6_dport;
907 	uh->len = htons(up->len);
908 	uh->check = 0;
909 
910 	if (is_udplite)
911 		csum = udplite_csum_outgoing(sk, skb);
912 	else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
913 		udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr,
914 				     up->len);
915 		goto send;
916 	} else
917 		csum = udp_csum_outgoing(sk, skb);
918 
919 	/* add protocol-dependent pseudo-header */
920 	uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
921 				    up->len, fl6->flowi6_proto, csum);
922 	if (uh->check == 0)
923 		uh->check = CSUM_MANGLED_0;
924 
925 send:
926 	err = ip6_push_pending_frames(sk);
927 	if (err) {
928 		if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
929 			UDP6_INC_STATS_USER(sock_net(sk),
930 					    UDP_MIB_SNDBUFERRORS, is_udplite);
931 			err = 0;
932 		}
933 	} else
934 		UDP6_INC_STATS_USER(sock_net(sk),
935 				    UDP_MIB_OUTDATAGRAMS, is_udplite);
936 out:
937 	up->len = 0;
938 	up->pending = 0;
939 	return err;
940 }
941 
942 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
943 		  struct msghdr *msg, size_t len)
944 {
945 	struct ipv6_txoptions opt_space;
946 	struct udp_sock *up = udp_sk(sk);
947 	struct inet_sock *inet = inet_sk(sk);
948 	struct ipv6_pinfo *np = inet6_sk(sk);
949 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
950 	struct in6_addr *daddr, *final_p, final;
951 	struct ipv6_txoptions *opt = NULL;
952 	struct ip6_flowlabel *flowlabel = NULL;
953 	struct flowi6 fl6;
954 	struct dst_entry *dst;
955 	int addr_len = msg->msg_namelen;
956 	int ulen = len;
957 	int hlimit = -1;
958 	int tclass = -1;
959 	int dontfrag = -1;
960 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
961 	int err;
962 	int connected = 0;
963 	int is_udplite = IS_UDPLITE(sk);
964 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
965 
966 	/* destination address check */
967 	if (sin6) {
968 		if (addr_len < offsetof(struct sockaddr, sa_data))
969 			return -EINVAL;
970 
971 		switch (sin6->sin6_family) {
972 		case AF_INET6:
973 			if (addr_len < SIN6_LEN_RFC2133)
974 				return -EINVAL;
975 			daddr = &sin6->sin6_addr;
976 			break;
977 		case AF_INET:
978 			goto do_udp_sendmsg;
979 		case AF_UNSPEC:
980 			msg->msg_name = sin6 = NULL;
981 			msg->msg_namelen = addr_len = 0;
982 			daddr = NULL;
983 			break;
984 		default:
985 			return -EINVAL;
986 		}
987 	} else if (!up->pending) {
988 		if (sk->sk_state != TCP_ESTABLISHED)
989 			return -EDESTADDRREQ;
990 		daddr = &np->daddr;
991 	} else
992 		daddr = NULL;
993 
994 	if (daddr) {
995 		if (ipv6_addr_v4mapped(daddr)) {
996 			struct sockaddr_in sin;
997 			sin.sin_family = AF_INET;
998 			sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
999 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
1000 			msg->msg_name = &sin;
1001 			msg->msg_namelen = sizeof(sin);
1002 do_udp_sendmsg:
1003 			if (__ipv6_only_sock(sk))
1004 				return -ENETUNREACH;
1005 			return udp_sendmsg(iocb, sk, msg, len);
1006 		}
1007 	}
1008 
1009 	if (up->pending == AF_INET)
1010 		return udp_sendmsg(iocb, sk, msg, len);
1011 
1012 	/* Rough check on arithmetic overflow,
1013 	   better check is made in ip6_append_data().
1014 	   */
1015 	if (len > INT_MAX - sizeof(struct udphdr))
1016 		return -EMSGSIZE;
1017 
1018 	if (up->pending) {
1019 		/*
1020 		 * There are pending frames.
1021 		 * The socket lock must be held while it's corked.
1022 		 */
1023 		lock_sock(sk);
1024 		if (likely(up->pending)) {
1025 			if (unlikely(up->pending != AF_INET6)) {
1026 				release_sock(sk);
1027 				return -EAFNOSUPPORT;
1028 			}
1029 			dst = NULL;
1030 			goto do_append_data;
1031 		}
1032 		release_sock(sk);
1033 	}
1034 	ulen += sizeof(struct udphdr);
1035 
1036 	memset(&fl6, 0, sizeof(fl6));
1037 
1038 	if (sin6) {
1039 		if (sin6->sin6_port == 0)
1040 			return -EINVAL;
1041 
1042 		fl6.fl6_dport = sin6->sin6_port;
1043 		daddr = &sin6->sin6_addr;
1044 
1045 		if (np->sndflow) {
1046 			fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1047 			if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1048 				flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1049 				if (flowlabel == NULL)
1050 					return -EINVAL;
1051 				daddr = &flowlabel->dst;
1052 			}
1053 		}
1054 
1055 		/*
1056 		 * Otherwise it will be difficult to maintain
1057 		 * sk->sk_dst_cache.
1058 		 */
1059 		if (sk->sk_state == TCP_ESTABLISHED &&
1060 		    ipv6_addr_equal(daddr, &np->daddr))
1061 			daddr = &np->daddr;
1062 
1063 		if (addr_len >= sizeof(struct sockaddr_in6) &&
1064 		    sin6->sin6_scope_id &&
1065 		    ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
1066 			fl6.flowi6_oif = sin6->sin6_scope_id;
1067 	} else {
1068 		if (sk->sk_state != TCP_ESTABLISHED)
1069 			return -EDESTADDRREQ;
1070 
1071 		fl6.fl6_dport = inet->inet_dport;
1072 		daddr = &np->daddr;
1073 		fl6.flowlabel = np->flow_label;
1074 		connected = 1;
1075 	}
1076 
1077 	if (!fl6.flowi6_oif)
1078 		fl6.flowi6_oif = sk->sk_bound_dev_if;
1079 
1080 	if (!fl6.flowi6_oif)
1081 		fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1082 
1083 	fl6.flowi6_mark = sk->sk_mark;
1084 
1085 	if (msg->msg_controllen) {
1086 		opt = &opt_space;
1087 		memset(opt, 0, sizeof(struct ipv6_txoptions));
1088 		opt->tot_len = sizeof(*opt);
1089 
1090 		err = datagram_send_ctl(sock_net(sk), msg, &fl6, opt, &hlimit,
1091 					&tclass, &dontfrag);
1092 		if (err < 0) {
1093 			fl6_sock_release(flowlabel);
1094 			return err;
1095 		}
1096 		if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1097 			flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1098 			if (flowlabel == NULL)
1099 				return -EINVAL;
1100 		}
1101 		if (!(opt->opt_nflen|opt->opt_flen))
1102 			opt = NULL;
1103 		connected = 0;
1104 	}
1105 	if (opt == NULL)
1106 		opt = np->opt;
1107 	if (flowlabel)
1108 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
1109 	opt = ipv6_fixup_options(&opt_space, opt);
1110 
1111 	fl6.flowi6_proto = sk->sk_protocol;
1112 	if (!ipv6_addr_any(daddr))
1113 		ipv6_addr_copy(&fl6.daddr, daddr);
1114 	else
1115 		fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1116 	if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1117 		ipv6_addr_copy(&fl6.saddr, &np->saddr);
1118 	fl6.fl6_sport = inet->inet_sport;
1119 
1120 	final_p = fl6_update_dst(&fl6, opt, &final);
1121 	if (final_p)
1122 		connected = 0;
1123 
1124 	if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1125 		fl6.flowi6_oif = np->mcast_oif;
1126 		connected = 0;
1127 	}
1128 
1129 	security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1130 
1131 	dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, true);
1132 	if (IS_ERR(dst)) {
1133 		err = PTR_ERR(dst);
1134 		dst = NULL;
1135 		goto out;
1136 	}
1137 
1138 	if (hlimit < 0) {
1139 		if (ipv6_addr_is_multicast(&fl6.daddr))
1140 			hlimit = np->mcast_hops;
1141 		else
1142 			hlimit = np->hop_limit;
1143 		if (hlimit < 0)
1144 			hlimit = ip6_dst_hoplimit(dst);
1145 	}
1146 
1147 	if (tclass < 0)
1148 		tclass = np->tclass;
1149 
1150 	if (dontfrag < 0)
1151 		dontfrag = np->dontfrag;
1152 
1153 	if (msg->msg_flags&MSG_CONFIRM)
1154 		goto do_confirm;
1155 back_from_confirm:
1156 
1157 	lock_sock(sk);
1158 	if (unlikely(up->pending)) {
1159 		/* The socket is already corked while preparing it. */
1160 		/* ... which is an evident application bug. --ANK */
1161 		release_sock(sk);
1162 
1163 		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1164 		err = -EINVAL;
1165 		goto out;
1166 	}
1167 
1168 	up->pending = AF_INET6;
1169 
1170 do_append_data:
1171 	up->len += ulen;
1172 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
1173 	err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1174 		sizeof(struct udphdr), hlimit, tclass, opt, &fl6,
1175 		(struct rt6_info*)dst,
1176 		corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1177 	if (err)
1178 		udp_v6_flush_pending_frames(sk);
1179 	else if (!corkreq)
1180 		err = udp_v6_push_pending_frames(sk);
1181 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1182 		up->pending = 0;
1183 
1184 	if (dst) {
1185 		if (connected) {
1186 			ip6_dst_store(sk, dst,
1187 				      ipv6_addr_equal(&fl6.daddr, &np->daddr) ?
1188 				      &np->daddr : NULL,
1189 #ifdef CONFIG_IPV6_SUBTREES
1190 				      ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
1191 				      &np->saddr :
1192 #endif
1193 				      NULL);
1194 		} else {
1195 			dst_release(dst);
1196 		}
1197 		dst = NULL;
1198 	}
1199 
1200 	if (err > 0)
1201 		err = np->recverr ? net_xmit_errno(err) : 0;
1202 	release_sock(sk);
1203 out:
1204 	dst_release(dst);
1205 	fl6_sock_release(flowlabel);
1206 	if (!err)
1207 		return len;
1208 	/*
1209 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1210 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1211 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1212 	 * things).  We could add another new stat but at least for now that
1213 	 * seems like overkill.
1214 	 */
1215 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1216 		UDP6_INC_STATS_USER(sock_net(sk),
1217 				UDP_MIB_SNDBUFERRORS, is_udplite);
1218 	}
1219 	return err;
1220 
1221 do_confirm:
1222 	dst_confirm(dst);
1223 	if (!(msg->msg_flags&MSG_PROBE) || len)
1224 		goto back_from_confirm;
1225 	err = 0;
1226 	goto out;
1227 }
1228 
1229 void udpv6_destroy_sock(struct sock *sk)
1230 {
1231 	lock_sock(sk);
1232 	udp_v6_flush_pending_frames(sk);
1233 	release_sock(sk);
1234 
1235 	inet6_destroy_sock(sk);
1236 }
1237 
1238 /*
1239  *	Socket option code for UDP
1240  */
1241 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1242 		     char __user *optval, unsigned int optlen)
1243 {
1244 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1245 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1246 					  udp_v6_push_pending_frames);
1247 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
1248 }
1249 
1250 #ifdef CONFIG_COMPAT
1251 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1252 			    char __user *optval, unsigned int optlen)
1253 {
1254 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1255 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1256 					  udp_v6_push_pending_frames);
1257 	return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1258 }
1259 #endif
1260 
1261 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1262 		     char __user *optval, int __user *optlen)
1263 {
1264 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1265 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1266 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
1267 }
1268 
1269 #ifdef CONFIG_COMPAT
1270 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1271 			    char __user *optval, int __user *optlen)
1272 {
1273 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1274 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1275 	return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1276 }
1277 #endif
1278 
1279 static int udp6_ufo_send_check(struct sk_buff *skb)
1280 {
1281 	const struct ipv6hdr *ipv6h;
1282 	struct udphdr *uh;
1283 
1284 	if (!pskb_may_pull(skb, sizeof(*uh)))
1285 		return -EINVAL;
1286 
1287 	ipv6h = ipv6_hdr(skb);
1288 	uh = udp_hdr(skb);
1289 
1290 	uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
1291 				     IPPROTO_UDP, 0);
1292 	skb->csum_start = skb_transport_header(skb) - skb->head;
1293 	skb->csum_offset = offsetof(struct udphdr, check);
1294 	skb->ip_summed = CHECKSUM_PARTIAL;
1295 	return 0;
1296 }
1297 
1298 static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb, u32 features)
1299 {
1300 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1301 	unsigned int mss;
1302 	unsigned int unfrag_ip6hlen, unfrag_len;
1303 	struct frag_hdr *fptr;
1304 	u8 *mac_start, *prevhdr;
1305 	u8 nexthdr;
1306 	u8 frag_hdr_sz = sizeof(struct frag_hdr);
1307 	int offset;
1308 	__wsum csum;
1309 
1310 	mss = skb_shinfo(skb)->gso_size;
1311 	if (unlikely(skb->len <= mss))
1312 		goto out;
1313 
1314 	if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1315 		/* Packet is from an untrusted source, reset gso_segs. */
1316 		int type = skb_shinfo(skb)->gso_type;
1317 
1318 		if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1319 			     !(type & (SKB_GSO_UDP))))
1320 			goto out;
1321 
1322 		skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1323 
1324 		segs = NULL;
1325 		goto out;
1326 	}
1327 
1328 	/* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1329 	 * do checksum of UDP packets sent as multiple IP fragments.
1330 	 */
1331 	offset = skb_checksum_start_offset(skb);
1332 	csum = skb_checksum(skb, offset, skb->len- offset, 0);
1333 	offset += skb->csum_offset;
1334 	*(__sum16 *)(skb->data + offset) = csum_fold(csum);
1335 	skb->ip_summed = CHECKSUM_NONE;
1336 
1337 	/* Check if there is enough headroom to insert fragment header. */
1338 	if ((skb_mac_header(skb) < skb->head + frag_hdr_sz) &&
1339 	    pskb_expand_head(skb, frag_hdr_sz, 0, GFP_ATOMIC))
1340 		goto out;
1341 
1342 	/* Find the unfragmentable header and shift it left by frag_hdr_sz
1343 	 * bytes to insert fragment header.
1344 	 */
1345 	unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
1346 	nexthdr = *prevhdr;
1347 	*prevhdr = NEXTHDR_FRAGMENT;
1348 	unfrag_len = skb_network_header(skb) - skb_mac_header(skb) +
1349 		     unfrag_ip6hlen;
1350 	mac_start = skb_mac_header(skb);
1351 	memmove(mac_start-frag_hdr_sz, mac_start, unfrag_len);
1352 
1353 	skb->mac_header -= frag_hdr_sz;
1354 	skb->network_header -= frag_hdr_sz;
1355 
1356 	fptr = (struct frag_hdr *)(skb_network_header(skb) + unfrag_ip6hlen);
1357 	fptr->nexthdr = nexthdr;
1358 	fptr->reserved = 0;
1359 	ipv6_select_ident(fptr);
1360 
1361 	/* Fragment the skb. ipv6 header and the remaining fields of the
1362 	 * fragment header are updated in ipv6_gso_segment()
1363 	 */
1364 	segs = skb_segment(skb, features);
1365 
1366 out:
1367 	return segs;
1368 }
1369 
1370 static const struct inet6_protocol udpv6_protocol = {
1371 	.handler	=	udpv6_rcv,
1372 	.err_handler	=	udpv6_err,
1373 	.gso_send_check =	udp6_ufo_send_check,
1374 	.gso_segment	=	udp6_ufo_fragment,
1375 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1376 };
1377 
1378 /* ------------------------------------------------------------------------ */
1379 #ifdef CONFIG_PROC_FS
1380 
1381 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
1382 {
1383 	struct inet_sock *inet = inet_sk(sp);
1384 	struct ipv6_pinfo *np = inet6_sk(sp);
1385 	const struct in6_addr *dest, *src;
1386 	__u16 destp, srcp;
1387 
1388 	dest  = &np->daddr;
1389 	src   = &np->rcv_saddr;
1390 	destp = ntohs(inet->inet_dport);
1391 	srcp  = ntohs(inet->inet_sport);
1392 	seq_printf(seq,
1393 		   "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1394 		   "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n",
1395 		   bucket,
1396 		   src->s6_addr32[0], src->s6_addr32[1],
1397 		   src->s6_addr32[2], src->s6_addr32[3], srcp,
1398 		   dest->s6_addr32[0], dest->s6_addr32[1],
1399 		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
1400 		   sp->sk_state,
1401 		   sk_wmem_alloc_get(sp),
1402 		   sk_rmem_alloc_get(sp),
1403 		   0, 0L, 0,
1404 		   sock_i_uid(sp), 0,
1405 		   sock_i_ino(sp),
1406 		   atomic_read(&sp->sk_refcnt), sp,
1407 		   atomic_read(&sp->sk_drops));
1408 }
1409 
1410 int udp6_seq_show(struct seq_file *seq, void *v)
1411 {
1412 	if (v == SEQ_START_TOKEN)
1413 		seq_printf(seq,
1414 			   "  sl  "
1415 			   "local_address                         "
1416 			   "remote_address                        "
1417 			   "st tx_queue rx_queue tr tm->when retrnsmt"
1418 			   "   uid  timeout inode ref pointer drops\n");
1419 	else
1420 		udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1421 	return 0;
1422 }
1423 
1424 static struct udp_seq_afinfo udp6_seq_afinfo = {
1425 	.name		= "udp6",
1426 	.family		= AF_INET6,
1427 	.udp_table	= &udp_table,
1428 	.seq_fops	= {
1429 		.owner	=	THIS_MODULE,
1430 	},
1431 	.seq_ops	= {
1432 		.show		= udp6_seq_show,
1433 	},
1434 };
1435 
1436 int __net_init udp6_proc_init(struct net *net)
1437 {
1438 	return udp_proc_register(net, &udp6_seq_afinfo);
1439 }
1440 
1441 void udp6_proc_exit(struct net *net) {
1442 	udp_proc_unregister(net, &udp6_seq_afinfo);
1443 }
1444 #endif /* CONFIG_PROC_FS */
1445 
1446 /* ------------------------------------------------------------------------ */
1447 
1448 struct proto udpv6_prot = {
1449 	.name		   = "UDPv6",
1450 	.owner		   = THIS_MODULE,
1451 	.close		   = udp_lib_close,
1452 	.connect	   = ip6_datagram_connect,
1453 	.disconnect	   = udp_disconnect,
1454 	.ioctl		   = udp_ioctl,
1455 	.destroy	   = udpv6_destroy_sock,
1456 	.setsockopt	   = udpv6_setsockopt,
1457 	.getsockopt	   = udpv6_getsockopt,
1458 	.sendmsg	   = udpv6_sendmsg,
1459 	.recvmsg	   = udpv6_recvmsg,
1460 	.backlog_rcv	   = udpv6_queue_rcv_skb,
1461 	.hash		   = udp_lib_hash,
1462 	.unhash		   = udp_lib_unhash,
1463 	.rehash		   = udp_v6_rehash,
1464 	.get_port	   = udp_v6_get_port,
1465 	.memory_allocated  = &udp_memory_allocated,
1466 	.sysctl_mem	   = sysctl_udp_mem,
1467 	.sysctl_wmem	   = &sysctl_udp_wmem_min,
1468 	.sysctl_rmem	   = &sysctl_udp_rmem_min,
1469 	.obj_size	   = sizeof(struct udp6_sock),
1470 	.slab_flags	   = SLAB_DESTROY_BY_RCU,
1471 	.h.udp_table	   = &udp_table,
1472 #ifdef CONFIG_COMPAT
1473 	.compat_setsockopt = compat_udpv6_setsockopt,
1474 	.compat_getsockopt = compat_udpv6_getsockopt,
1475 #endif
1476 	.clear_sk	   = sk_prot_clear_portaddr_nulls,
1477 };
1478 
1479 static struct inet_protosw udpv6_protosw = {
1480 	.type =      SOCK_DGRAM,
1481 	.protocol =  IPPROTO_UDP,
1482 	.prot =      &udpv6_prot,
1483 	.ops =       &inet6_dgram_ops,
1484 	.no_check =  UDP_CSUM_DEFAULT,
1485 	.flags =     INET_PROTOSW_PERMANENT,
1486 };
1487 
1488 
1489 int __init udpv6_init(void)
1490 {
1491 	int ret;
1492 
1493 	ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1494 	if (ret)
1495 		goto out;
1496 
1497 	ret = inet6_register_protosw(&udpv6_protosw);
1498 	if (ret)
1499 		goto out_udpv6_protocol;
1500 out:
1501 	return ret;
1502 
1503 out_udpv6_protocol:
1504 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1505 	goto out;
1506 }
1507 
1508 void udpv6_exit(void)
1509 {
1510 	inet6_unregister_protosw(&udpv6_protosw);
1511 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1512 }
1513