xref: /linux/net/ipv6/udp.c (revision a33f32244d8550da8b4a26e277ce07d5c6d158b5)
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 = inet_sk(sk)->inet_rcv_saddr;
58 	__be32 sk2_rcv_saddr = inet_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(addr6->s6_addr32[3], mix);
95 	else
96 		hash = jhash2(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 inline int compute_score(struct sock *sk, struct net *net,
115 				unsigned short hnum,
116 				struct in6_addr *saddr, __be16 sport,
117 				struct in6_addr *daddr, __be16 dport,
118 				int dif)
119 {
120 	int score = -1;
121 
122 	if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
123 			sk->sk_family == PF_INET6) {
124 		struct ipv6_pinfo *np = inet6_sk(sk);
125 		struct inet_sock *inet = inet_sk(sk);
126 
127 		score = 0;
128 		if (inet->inet_dport) {
129 			if (inet->inet_dport != sport)
130 				return -1;
131 			score++;
132 		}
133 		if (!ipv6_addr_any(&np->rcv_saddr)) {
134 			if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
135 				return -1;
136 			score++;
137 		}
138 		if (!ipv6_addr_any(&np->daddr)) {
139 			if (!ipv6_addr_equal(&np->daddr, saddr))
140 				return -1;
141 			score++;
142 		}
143 		if (sk->sk_bound_dev_if) {
144 			if (sk->sk_bound_dev_if != dif)
145 				return -1;
146 			score++;
147 		}
148 	}
149 	return score;
150 }
151 
152 #define SCORE2_MAX (1 + 1 + 1)
153 static inline int compute_score2(struct sock *sk, struct net *net,
154 				const struct in6_addr *saddr, __be16 sport,
155 				const struct in6_addr *daddr, unsigned short hnum,
156 				int dif)
157 {
158 	int score = -1;
159 
160 	if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
161 			sk->sk_family == PF_INET6) {
162 		struct ipv6_pinfo *np = inet6_sk(sk);
163 		struct inet_sock *inet = inet_sk(sk);
164 
165 		if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
166 			return -1;
167 		score = 0;
168 		if (inet->inet_dport) {
169 			if (inet->inet_dport != sport)
170 				return -1;
171 			score++;
172 		}
173 		if (!ipv6_addr_any(&np->daddr)) {
174 			if (!ipv6_addr_equal(&np->daddr, saddr))
175 				return -1;
176 			score++;
177 		}
178 		if (sk->sk_bound_dev_if) {
179 			if (sk->sk_bound_dev_if != dif)
180 				return -1;
181 			score++;
182 		}
183 	}
184 	return score;
185 }
186 
187 
188 /* called with read_rcu_lock() */
189 static struct sock *udp6_lib_lookup2(struct net *net,
190 		const struct in6_addr *saddr, __be16 sport,
191 		const struct in6_addr *daddr, unsigned int hnum, int dif,
192 		struct udp_hslot *hslot2, unsigned int slot2)
193 {
194 	struct sock *sk, *result;
195 	struct hlist_nulls_node *node;
196 	int score, badness;
197 
198 begin:
199 	result = NULL;
200 	badness = -1;
201 	udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
202 		score = compute_score2(sk, net, saddr, sport,
203 				      daddr, hnum, dif);
204 		if (score > badness) {
205 			result = sk;
206 			badness = score;
207 			if (score == SCORE2_MAX)
208 				goto exact_match;
209 		}
210 	}
211 	/*
212 	 * if the nulls value we got at the end of this lookup is
213 	 * not the expected one, we must restart lookup.
214 	 * We probably met an item that was moved to another chain.
215 	 */
216 	if (get_nulls_value(node) != slot2)
217 		goto begin;
218 
219 	if (result) {
220 exact_match:
221 		if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
222 			result = NULL;
223 		else if (unlikely(compute_score2(result, net, saddr, sport,
224 				  daddr, hnum, dif) < badness)) {
225 			sock_put(result);
226 			goto begin;
227 		}
228 	}
229 	return result;
230 }
231 
232 static struct sock *__udp6_lib_lookup(struct net *net,
233 				      struct in6_addr *saddr, __be16 sport,
234 				      struct in6_addr *daddr, __be16 dport,
235 				      int dif, struct udp_table *udptable)
236 {
237 	struct sock *sk, *result;
238 	struct hlist_nulls_node *node;
239 	unsigned short hnum = ntohs(dport);
240 	unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
241 	struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
242 	int score, badness;
243 
244 	rcu_read_lock();
245 	if (hslot->count > 10) {
246 		hash2 = udp6_portaddr_hash(net, daddr, hnum);
247 		slot2 = hash2 & udptable->mask;
248 		hslot2 = &udptable->hash2[slot2];
249 		if (hslot->count < hslot2->count)
250 			goto begin;
251 
252 		result = udp6_lib_lookup2(net, saddr, sport,
253 					  daddr, hnum, dif,
254 					  hslot2, slot2);
255 		if (!result) {
256 			hash2 = udp6_portaddr_hash(net, &in6addr_any, 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 						  &in6addr_any, hnum, dif,
264 						  hslot2, slot2);
265 		}
266 		rcu_read_unlock();
267 		return result;
268 	}
269 begin:
270 	result = NULL;
271 	badness = -1;
272 	sk_nulls_for_each_rcu(sk, node, &hslot->head) {
273 		score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif);
274 		if (score > badness) {
275 			result = sk;
276 			badness = score;
277 		}
278 	}
279 	/*
280 	 * if the nulls value we got at the end of this lookup is
281 	 * not the expected one, we must restart lookup.
282 	 * We probably met an item that was moved to another chain.
283 	 */
284 	if (get_nulls_value(node) != slot)
285 		goto begin;
286 
287 	if (result) {
288 		if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
289 			result = NULL;
290 		else if (unlikely(compute_score(result, net, hnum, saddr, sport,
291 					daddr, dport, dif) < badness)) {
292 			sock_put(result);
293 			goto begin;
294 		}
295 	}
296 	rcu_read_unlock();
297 	return result;
298 }
299 
300 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
301 					  __be16 sport, __be16 dport,
302 					  struct udp_table *udptable)
303 {
304 	struct sock *sk;
305 	struct ipv6hdr *iph = ipv6_hdr(skb);
306 
307 	if (unlikely(sk = skb_steal_sock(skb)))
308 		return sk;
309 	return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport,
310 				 &iph->daddr, dport, inet6_iif(skb),
311 				 udptable);
312 }
313 
314 /*
315  * 	This should be easy, if there is something there we
316  * 	return it, otherwise we block.
317  */
318 
319 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
320 		  struct msghdr *msg, size_t len,
321 		  int noblock, int flags, int *addr_len)
322 {
323 	struct ipv6_pinfo *np = inet6_sk(sk);
324 	struct inet_sock *inet = inet_sk(sk);
325 	struct sk_buff *skb;
326 	unsigned int ulen;
327 	int peeked;
328 	int err;
329 	int is_udplite = IS_UDPLITE(sk);
330 	int is_udp4;
331 
332 	if (addr_len)
333 		*addr_len=sizeof(struct sockaddr_in6);
334 
335 	if (flags & MSG_ERRQUEUE)
336 		return ipv6_recv_error(sk, msg, len);
337 
338 try_again:
339 	skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
340 				  &peeked, &err);
341 	if (!skb)
342 		goto out;
343 
344 	ulen = skb->len - sizeof(struct udphdr);
345 	if (len > ulen)
346 		len = ulen;
347 	else if (len < ulen)
348 		msg->msg_flags |= MSG_TRUNC;
349 
350 	is_udp4 = (skb->protocol == htons(ETH_P_IP));
351 
352 	/*
353 	 * If checksum is needed at all, try to do it while copying the
354 	 * data.  If the data is truncated, or if we only want a partial
355 	 * coverage checksum (UDP-Lite), do it before the copy.
356 	 */
357 
358 	if (len < ulen || UDP_SKB_CB(skb)->partial_cov) {
359 		if (udp_lib_checksum_complete(skb))
360 			goto csum_copy_err;
361 	}
362 
363 	if (skb_csum_unnecessary(skb))
364 		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
365 					      msg->msg_iov,len);
366 	else {
367 		err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
368 		if (err == -EINVAL)
369 			goto csum_copy_err;
370 	}
371 	if (err)
372 		goto out_free;
373 
374 	if (!peeked) {
375 		if (is_udp4)
376 			UDP_INC_STATS_USER(sock_net(sk),
377 					UDP_MIB_INDATAGRAMS, is_udplite);
378 		else
379 			UDP6_INC_STATS_USER(sock_net(sk),
380 					UDP_MIB_INDATAGRAMS, is_udplite);
381 	}
382 
383 	sock_recv_ts_and_drops(msg, sk, skb);
384 
385 	/* Copy the address. */
386 	if (msg->msg_name) {
387 		struct sockaddr_in6 *sin6;
388 
389 		sin6 = (struct sockaddr_in6 *) msg->msg_name;
390 		sin6->sin6_family = AF_INET6;
391 		sin6->sin6_port = udp_hdr(skb)->source;
392 		sin6->sin6_flowinfo = 0;
393 		sin6->sin6_scope_id = 0;
394 
395 		if (is_udp4)
396 			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
397 					       &sin6->sin6_addr);
398 		else {
399 			ipv6_addr_copy(&sin6->sin6_addr,
400 				       &ipv6_hdr(skb)->saddr);
401 			if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
402 				sin6->sin6_scope_id = IP6CB(skb)->iif;
403 		}
404 
405 	}
406 	if (is_udp4) {
407 		if (inet->cmsg_flags)
408 			ip_cmsg_recv(msg, skb);
409 	} else {
410 		if (np->rxopt.all)
411 			datagram_recv_ctl(sk, msg, skb);
412 	}
413 
414 	err = len;
415 	if (flags & MSG_TRUNC)
416 		err = ulen;
417 
418 out_free:
419 	skb_free_datagram_locked(sk, skb);
420 out:
421 	return err;
422 
423 csum_copy_err:
424 	lock_sock(sk);
425 	if (!skb_kill_datagram(sk, skb, flags)) {
426 		if (is_udp4)
427 			UDP_INC_STATS_USER(sock_net(sk),
428 					UDP_MIB_INERRORS, is_udplite);
429 		else
430 			UDP6_INC_STATS_USER(sock_net(sk),
431 					UDP_MIB_INERRORS, is_udplite);
432 	}
433 	release_sock(sk);
434 
435 	if (flags & MSG_DONTWAIT)
436 		return -EAGAIN;
437 	goto try_again;
438 }
439 
440 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
441 		    u8 type, u8 code, int offset, __be32 info,
442 		    struct udp_table *udptable)
443 {
444 	struct ipv6_pinfo *np;
445 	struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
446 	struct in6_addr *saddr = &hdr->saddr;
447 	struct in6_addr *daddr = &hdr->daddr;
448 	struct udphdr *uh = (struct udphdr*)(skb->data+offset);
449 	struct sock *sk;
450 	int err;
451 
452 	sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
453 			       saddr, uh->source, inet6_iif(skb), udptable);
454 	if (sk == NULL)
455 		return;
456 
457 	np = inet6_sk(sk);
458 
459 	if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
460 		goto out;
461 
462 	if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
463 		goto out;
464 
465 	if (np->recverr)
466 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
467 
468 	sk->sk_err = err;
469 	sk->sk_error_report(sk);
470 out:
471 	sock_put(sk);
472 }
473 
474 static __inline__ void udpv6_err(struct sk_buff *skb,
475 				 struct inet6_skb_parm *opt, u8 type,
476 				 u8 code, int offset, __be32 info     )
477 {
478 	__udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
479 }
480 
481 int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
482 {
483 	struct udp_sock *up = udp_sk(sk);
484 	int rc;
485 	int is_udplite = IS_UDPLITE(sk);
486 
487 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
488 		goto drop;
489 
490 	/*
491 	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
492 	 */
493 	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
494 
495 		if (up->pcrlen == 0) {          /* full coverage was set  */
496 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
497 				" %d while full coverage %d requested\n",
498 				UDP_SKB_CB(skb)->cscov, skb->len);
499 			goto drop;
500 		}
501 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
502 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
503 						    "too small, need min %d\n",
504 				       UDP_SKB_CB(skb)->cscov, up->pcrlen);
505 			goto drop;
506 		}
507 	}
508 
509 	if (sk->sk_filter) {
510 		if (udp_lib_checksum_complete(skb))
511 			goto drop;
512 	}
513 
514 	if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) {
515 		/* Note that an ENOMEM error is charged twice */
516 		if (rc == -ENOMEM)
517 			UDP6_INC_STATS_BH(sock_net(sk),
518 					UDP_MIB_RCVBUFERRORS, is_udplite);
519 		goto drop_no_sk_drops_inc;
520 	}
521 
522 	return 0;
523 drop:
524 	atomic_inc(&sk->sk_drops);
525 drop_no_sk_drops_inc:
526 	UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
527 	kfree_skb(skb);
528 	return -1;
529 }
530 
531 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
532 				      __be16 loc_port, struct in6_addr *loc_addr,
533 				      __be16 rmt_port, struct in6_addr *rmt_addr,
534 				      int dif)
535 {
536 	struct hlist_nulls_node *node;
537 	struct sock *s = sk;
538 	unsigned short num = ntohs(loc_port);
539 
540 	sk_nulls_for_each_from(s, node) {
541 		struct inet_sock *inet = inet_sk(s);
542 
543 		if (!net_eq(sock_net(s), net))
544 			continue;
545 
546 		if (udp_sk(s)->udp_port_hash == num &&
547 		    s->sk_family == PF_INET6) {
548 			struct ipv6_pinfo *np = inet6_sk(s);
549 			if (inet->inet_dport) {
550 				if (inet->inet_dport != rmt_port)
551 					continue;
552 			}
553 			if (!ipv6_addr_any(&np->daddr) &&
554 			    !ipv6_addr_equal(&np->daddr, rmt_addr))
555 				continue;
556 
557 			if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
558 				continue;
559 
560 			if (!ipv6_addr_any(&np->rcv_saddr)) {
561 				if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
562 					continue;
563 			}
564 			if (!inet6_mc_check(s, loc_addr, rmt_addr))
565 				continue;
566 			return s;
567 		}
568 	}
569 	return NULL;
570 }
571 
572 static void flush_stack(struct sock **stack, unsigned int count,
573 			struct sk_buff *skb, unsigned int final)
574 {
575 	unsigned int i;
576 	struct sock *sk;
577 	struct sk_buff *skb1;
578 
579 	for (i = 0; i < count; i++) {
580 		skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
581 
582 		sk = stack[i];
583 		if (skb1) {
584 			bh_lock_sock(sk);
585 			if (!sock_owned_by_user(sk))
586 				udpv6_queue_rcv_skb(sk, skb1);
587 			else if (sk_add_backlog(sk, skb1)) {
588 				kfree_skb(skb1);
589 				bh_unlock_sock(sk);
590 				goto drop;
591 			}
592 			bh_unlock_sock(sk);
593 			continue;
594 		}
595 drop:
596 		atomic_inc(&sk->sk_drops);
597 		UDP6_INC_STATS_BH(sock_net(sk),
598 				UDP_MIB_RCVBUFERRORS, IS_UDPLITE(sk));
599 		UDP6_INC_STATS_BH(sock_net(sk),
600 				UDP_MIB_INERRORS, IS_UDPLITE(sk));
601 	}
602 }
603 /*
604  * Note: called only from the BH handler context,
605  * so we don't need to lock the hashes.
606  */
607 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
608 		struct in6_addr *saddr, struct in6_addr *daddr,
609 		struct udp_table *udptable)
610 {
611 	struct sock *sk, *stack[256 / sizeof(struct sock *)];
612 	const struct udphdr *uh = udp_hdr(skb);
613 	struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
614 	int dif;
615 	unsigned int i, count = 0;
616 
617 	spin_lock(&hslot->lock);
618 	sk = sk_nulls_head(&hslot->head);
619 	dif = inet6_iif(skb);
620 	sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
621 	while (sk) {
622 		stack[count++] = sk;
623 		sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
624 				       uh->source, saddr, dif);
625 		if (unlikely(count == ARRAY_SIZE(stack))) {
626 			if (!sk)
627 				break;
628 			flush_stack(stack, count, skb, ~0);
629 			count = 0;
630 		}
631 	}
632 	/*
633 	 * before releasing the lock, we must take reference on sockets
634 	 */
635 	for (i = 0; i < count; i++)
636 		sock_hold(stack[i]);
637 
638 	spin_unlock(&hslot->lock);
639 
640 	if (count) {
641 		flush_stack(stack, count, skb, count - 1);
642 
643 		for (i = 0; i < count; i++)
644 			sock_put(stack[i]);
645 	} else {
646 		kfree_skb(skb);
647 	}
648 	return 0;
649 }
650 
651 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
652 				 int proto)
653 {
654 	int err;
655 
656 	UDP_SKB_CB(skb)->partial_cov = 0;
657 	UDP_SKB_CB(skb)->cscov = skb->len;
658 
659 	if (proto == IPPROTO_UDPLITE) {
660 		err = udplite_checksum_init(skb, uh);
661 		if (err)
662 			return err;
663 	}
664 
665 	if (uh->check == 0) {
666 		/* RFC 2460 section 8.1 says that we SHOULD log
667 		   this error. Well, it is reasonable.
668 		 */
669 		LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
670 		return 1;
671 	}
672 	if (skb->ip_summed == CHECKSUM_COMPLETE &&
673 	    !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
674 			     skb->len, proto, skb->csum))
675 		skb->ip_summed = CHECKSUM_UNNECESSARY;
676 
677 	if (!skb_csum_unnecessary(skb))
678 		skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
679 							 &ipv6_hdr(skb)->daddr,
680 							 skb->len, proto, 0));
681 
682 	return 0;
683 }
684 
685 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
686 		   int proto)
687 {
688 	struct net *net = dev_net(skb->dev);
689 	struct sock *sk;
690 	struct udphdr *uh;
691 	struct in6_addr *saddr, *daddr;
692 	u32 ulen = 0;
693 
694 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
695 		goto short_packet;
696 
697 	saddr = &ipv6_hdr(skb)->saddr;
698 	daddr = &ipv6_hdr(skb)->daddr;
699 	uh = udp_hdr(skb);
700 
701 	ulen = ntohs(uh->len);
702 	if (ulen > skb->len)
703 		goto short_packet;
704 
705 	if (proto == IPPROTO_UDP) {
706 		/* UDP validates ulen. */
707 
708 		/* Check for jumbo payload */
709 		if (ulen == 0)
710 			ulen = skb->len;
711 
712 		if (ulen < sizeof(*uh))
713 			goto short_packet;
714 
715 		if (ulen < skb->len) {
716 			if (pskb_trim_rcsum(skb, ulen))
717 				goto short_packet;
718 			saddr = &ipv6_hdr(skb)->saddr;
719 			daddr = &ipv6_hdr(skb)->daddr;
720 			uh = udp_hdr(skb);
721 		}
722 	}
723 
724 	if (udp6_csum_init(skb, uh, proto))
725 		goto discard;
726 
727 	/*
728 	 *	Multicast receive code
729 	 */
730 	if (ipv6_addr_is_multicast(daddr))
731 		return __udp6_lib_mcast_deliver(net, skb,
732 				saddr, daddr, udptable);
733 
734 	/* Unicast */
735 
736 	/*
737 	 * check socket cache ... must talk to Alan about his plans
738 	 * for sock caches... i'll skip this for now.
739 	 */
740 	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
741 
742 	if (sk == NULL) {
743 		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
744 			goto discard;
745 
746 		if (udp_lib_checksum_complete(skb))
747 			goto discard;
748 		UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS,
749 				proto == IPPROTO_UDPLITE);
750 
751 		icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
752 
753 		kfree_skb(skb);
754 		return 0;
755 	}
756 
757 	/* deliver */
758 
759 	bh_lock_sock(sk);
760 	if (!sock_owned_by_user(sk))
761 		udpv6_queue_rcv_skb(sk, skb);
762 	else if (sk_add_backlog(sk, skb)) {
763 		atomic_inc(&sk->sk_drops);
764 		bh_unlock_sock(sk);
765 		sock_put(sk);
766 		goto discard;
767 	}
768 	bh_unlock_sock(sk);
769 	sock_put(sk);
770 	return 0;
771 
772 short_packet:
773 	LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: %d/%u\n",
774 		       proto == IPPROTO_UDPLITE ? "-Lite" : "",
775 		       ulen, skb->len);
776 
777 discard:
778 	UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
779 	kfree_skb(skb);
780 	return 0;
781 }
782 
783 static __inline__ int udpv6_rcv(struct sk_buff *skb)
784 {
785 	return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
786 }
787 
788 /*
789  * Throw away all pending data and cancel the corking. Socket is locked.
790  */
791 static void udp_v6_flush_pending_frames(struct sock *sk)
792 {
793 	struct udp_sock *up = udp_sk(sk);
794 
795 	if (up->pending == AF_INET)
796 		udp_flush_pending_frames(sk);
797 	else if (up->pending) {
798 		up->len = 0;
799 		up->pending = 0;
800 		ip6_flush_pending_frames(sk);
801 	}
802 }
803 
804 /**
805  * 	udp6_hwcsum_outgoing  -  handle outgoing HW checksumming
806  * 	@sk: 	socket we are sending on
807  * 	@skb: 	sk_buff containing the filled-in UDP header
808  * 	        (checksum field must be zeroed out)
809  */
810 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
811 				 const struct in6_addr *saddr,
812 				 const struct in6_addr *daddr, int len)
813 {
814 	unsigned int offset;
815 	struct udphdr *uh = udp_hdr(skb);
816 	__wsum csum = 0;
817 
818 	if (skb_queue_len(&sk->sk_write_queue) == 1) {
819 		/* Only one fragment on the socket.  */
820 		skb->csum_start = skb_transport_header(skb) - skb->head;
821 		skb->csum_offset = offsetof(struct udphdr, check);
822 		uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
823 	} else {
824 		/*
825 		 * HW-checksum won't work as there are two or more
826 		 * fragments on the socket so that all csums of sk_buffs
827 		 * should be together
828 		 */
829 		offset = skb_transport_offset(skb);
830 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
831 
832 		skb->ip_summed = CHECKSUM_NONE;
833 
834 		skb_queue_walk(&sk->sk_write_queue, skb) {
835 			csum = csum_add(csum, skb->csum);
836 		}
837 
838 		uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
839 					    csum);
840 		if (uh->check == 0)
841 			uh->check = CSUM_MANGLED_0;
842 	}
843 }
844 
845 /*
846  *	Sending
847  */
848 
849 static int udp_v6_push_pending_frames(struct sock *sk)
850 {
851 	struct sk_buff *skb;
852 	struct udphdr *uh;
853 	struct udp_sock  *up = udp_sk(sk);
854 	struct inet_sock *inet = inet_sk(sk);
855 	struct flowi *fl = &inet->cork.fl;
856 	int err = 0;
857 	int is_udplite = IS_UDPLITE(sk);
858 	__wsum csum = 0;
859 
860 	/* Grab the skbuff where UDP header space exists. */
861 	if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
862 		goto out;
863 
864 	/*
865 	 * Create a UDP header
866 	 */
867 	uh = udp_hdr(skb);
868 	uh->source = fl->fl_ip_sport;
869 	uh->dest = fl->fl_ip_dport;
870 	uh->len = htons(up->len);
871 	uh->check = 0;
872 
873 	if (is_udplite)
874 		csum = udplite_csum_outgoing(sk, skb);
875 	else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
876 		udp6_hwcsum_outgoing(sk, skb, &fl->fl6_src, &fl->fl6_dst,
877 				     up->len);
878 		goto send;
879 	} else
880 		csum = udp_csum_outgoing(sk, skb);
881 
882 	/* add protocol-dependent pseudo-header */
883 	uh->check = csum_ipv6_magic(&fl->fl6_src, &fl->fl6_dst,
884 				    up->len, fl->proto, csum   );
885 	if (uh->check == 0)
886 		uh->check = CSUM_MANGLED_0;
887 
888 send:
889 	err = ip6_push_pending_frames(sk);
890 	if (err) {
891 		if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
892 			UDP6_INC_STATS_USER(sock_net(sk),
893 					    UDP_MIB_SNDBUFERRORS, is_udplite);
894 			err = 0;
895 		}
896 	} else
897 		UDP6_INC_STATS_USER(sock_net(sk),
898 				    UDP_MIB_OUTDATAGRAMS, is_udplite);
899 out:
900 	up->len = 0;
901 	up->pending = 0;
902 	return err;
903 }
904 
905 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
906 		  struct msghdr *msg, size_t len)
907 {
908 	struct ipv6_txoptions opt_space;
909 	struct udp_sock *up = udp_sk(sk);
910 	struct inet_sock *inet = inet_sk(sk);
911 	struct ipv6_pinfo *np = inet6_sk(sk);
912 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
913 	struct in6_addr *daddr, *final_p = NULL, final;
914 	struct ipv6_txoptions *opt = NULL;
915 	struct ip6_flowlabel *flowlabel = NULL;
916 	struct flowi fl;
917 	struct dst_entry *dst;
918 	int addr_len = msg->msg_namelen;
919 	int ulen = len;
920 	int hlimit = -1;
921 	int tclass = -1;
922 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
923 	int err;
924 	int connected = 0;
925 	int is_udplite = IS_UDPLITE(sk);
926 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
927 
928 	/* destination address check */
929 	if (sin6) {
930 		if (addr_len < offsetof(struct sockaddr, sa_data))
931 			return -EINVAL;
932 
933 		switch (sin6->sin6_family) {
934 		case AF_INET6:
935 			if (addr_len < SIN6_LEN_RFC2133)
936 				return -EINVAL;
937 			daddr = &sin6->sin6_addr;
938 			break;
939 		case AF_INET:
940 			goto do_udp_sendmsg;
941 		case AF_UNSPEC:
942 			msg->msg_name = sin6 = NULL;
943 			msg->msg_namelen = addr_len = 0;
944 			daddr = NULL;
945 			break;
946 		default:
947 			return -EINVAL;
948 		}
949 	} else if (!up->pending) {
950 		if (sk->sk_state != TCP_ESTABLISHED)
951 			return -EDESTADDRREQ;
952 		daddr = &np->daddr;
953 	} else
954 		daddr = NULL;
955 
956 	if (daddr) {
957 		if (ipv6_addr_v4mapped(daddr)) {
958 			struct sockaddr_in sin;
959 			sin.sin_family = AF_INET;
960 			sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
961 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
962 			msg->msg_name = &sin;
963 			msg->msg_namelen = sizeof(sin);
964 do_udp_sendmsg:
965 			if (__ipv6_only_sock(sk))
966 				return -ENETUNREACH;
967 			return udp_sendmsg(iocb, sk, msg, len);
968 		}
969 	}
970 
971 	if (up->pending == AF_INET)
972 		return udp_sendmsg(iocb, sk, msg, len);
973 
974 	/* Rough check on arithmetic overflow,
975 	   better check is made in ip6_append_data().
976 	   */
977 	if (len > INT_MAX - sizeof(struct udphdr))
978 		return -EMSGSIZE;
979 
980 	if (up->pending) {
981 		/*
982 		 * There are pending frames.
983 		 * The socket lock must be held while it's corked.
984 		 */
985 		lock_sock(sk);
986 		if (likely(up->pending)) {
987 			if (unlikely(up->pending != AF_INET6)) {
988 				release_sock(sk);
989 				return -EAFNOSUPPORT;
990 			}
991 			dst = NULL;
992 			goto do_append_data;
993 		}
994 		release_sock(sk);
995 	}
996 	ulen += sizeof(struct udphdr);
997 
998 	memset(&fl, 0, sizeof(fl));
999 
1000 	if (sin6) {
1001 		if (sin6->sin6_port == 0)
1002 			return -EINVAL;
1003 
1004 		fl.fl_ip_dport = sin6->sin6_port;
1005 		daddr = &sin6->sin6_addr;
1006 
1007 		if (np->sndflow) {
1008 			fl.fl6_flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1009 			if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
1010 				flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
1011 				if (flowlabel == NULL)
1012 					return -EINVAL;
1013 				daddr = &flowlabel->dst;
1014 			}
1015 		}
1016 
1017 		/*
1018 		 * Otherwise it will be difficult to maintain
1019 		 * sk->sk_dst_cache.
1020 		 */
1021 		if (sk->sk_state == TCP_ESTABLISHED &&
1022 		    ipv6_addr_equal(daddr, &np->daddr))
1023 			daddr = &np->daddr;
1024 
1025 		if (addr_len >= sizeof(struct sockaddr_in6) &&
1026 		    sin6->sin6_scope_id &&
1027 		    ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
1028 			fl.oif = sin6->sin6_scope_id;
1029 	} else {
1030 		if (sk->sk_state != TCP_ESTABLISHED)
1031 			return -EDESTADDRREQ;
1032 
1033 		fl.fl_ip_dport = inet->inet_dport;
1034 		daddr = &np->daddr;
1035 		fl.fl6_flowlabel = np->flow_label;
1036 		connected = 1;
1037 	}
1038 
1039 	if (!fl.oif)
1040 		fl.oif = sk->sk_bound_dev_if;
1041 
1042 	if (!fl.oif)
1043 		fl.oif = np->sticky_pktinfo.ipi6_ifindex;
1044 
1045 	fl.mark = sk->sk_mark;
1046 
1047 	if (msg->msg_controllen) {
1048 		opt = &opt_space;
1049 		memset(opt, 0, sizeof(struct ipv6_txoptions));
1050 		opt->tot_len = sizeof(*opt);
1051 
1052 		err = datagram_send_ctl(sock_net(sk), msg, &fl, opt, &hlimit, &tclass);
1053 		if (err < 0) {
1054 			fl6_sock_release(flowlabel);
1055 			return err;
1056 		}
1057 		if ((fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1058 			flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
1059 			if (flowlabel == NULL)
1060 				return -EINVAL;
1061 		}
1062 		if (!(opt->opt_nflen|opt->opt_flen))
1063 			opt = NULL;
1064 		connected = 0;
1065 	}
1066 	if (opt == NULL)
1067 		opt = np->opt;
1068 	if (flowlabel)
1069 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
1070 	opt = ipv6_fixup_options(&opt_space, opt);
1071 
1072 	fl.proto = sk->sk_protocol;
1073 	if (!ipv6_addr_any(daddr))
1074 		ipv6_addr_copy(&fl.fl6_dst, daddr);
1075 	else
1076 		fl.fl6_dst.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1077 	if (ipv6_addr_any(&fl.fl6_src) && !ipv6_addr_any(&np->saddr))
1078 		ipv6_addr_copy(&fl.fl6_src, &np->saddr);
1079 	fl.fl_ip_sport = inet->inet_sport;
1080 
1081 	/* merge ip6_build_xmit from ip6_output */
1082 	if (opt && opt->srcrt) {
1083 		struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1084 		ipv6_addr_copy(&final, &fl.fl6_dst);
1085 		ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1086 		final_p = &final;
1087 		connected = 0;
1088 	}
1089 
1090 	if (!fl.oif && ipv6_addr_is_multicast(&fl.fl6_dst)) {
1091 		fl.oif = np->mcast_oif;
1092 		connected = 0;
1093 	}
1094 
1095 	security_sk_classify_flow(sk, &fl);
1096 
1097 	err = ip6_sk_dst_lookup(sk, &dst, &fl);
1098 	if (err)
1099 		goto out;
1100 	if (final_p)
1101 		ipv6_addr_copy(&fl.fl6_dst, final_p);
1102 
1103 	err = __xfrm_lookup(sock_net(sk), &dst, &fl, sk, XFRM_LOOKUP_WAIT);
1104 	if (err < 0) {
1105 		if (err == -EREMOTE)
1106 			err = ip6_dst_blackhole(sk, &dst, &fl);
1107 		if (err < 0)
1108 			goto out;
1109 	}
1110 
1111 	if (hlimit < 0) {
1112 		if (ipv6_addr_is_multicast(&fl.fl6_dst))
1113 			hlimit = np->mcast_hops;
1114 		else
1115 			hlimit = np->hop_limit;
1116 		if (hlimit < 0)
1117 			hlimit = ip6_dst_hoplimit(dst);
1118 	}
1119 
1120 	if (tclass < 0)
1121 		tclass = np->tclass;
1122 
1123 	if (msg->msg_flags&MSG_CONFIRM)
1124 		goto do_confirm;
1125 back_from_confirm:
1126 
1127 	lock_sock(sk);
1128 	if (unlikely(up->pending)) {
1129 		/* The socket is already corked while preparing it. */
1130 		/* ... which is an evident application bug. --ANK */
1131 		release_sock(sk);
1132 
1133 		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1134 		err = -EINVAL;
1135 		goto out;
1136 	}
1137 
1138 	up->pending = AF_INET6;
1139 
1140 do_append_data:
1141 	up->len += ulen;
1142 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
1143 	err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1144 		sizeof(struct udphdr), hlimit, tclass, opt, &fl,
1145 		(struct rt6_info*)dst,
1146 		corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1147 	if (err)
1148 		udp_v6_flush_pending_frames(sk);
1149 	else if (!corkreq)
1150 		err = udp_v6_push_pending_frames(sk);
1151 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1152 		up->pending = 0;
1153 
1154 	if (dst) {
1155 		if (connected) {
1156 			ip6_dst_store(sk, dst,
1157 				      ipv6_addr_equal(&fl.fl6_dst, &np->daddr) ?
1158 				      &np->daddr : NULL,
1159 #ifdef CONFIG_IPV6_SUBTREES
1160 				      ipv6_addr_equal(&fl.fl6_src, &np->saddr) ?
1161 				      &np->saddr :
1162 #endif
1163 				      NULL);
1164 		} else {
1165 			dst_release(dst);
1166 		}
1167 		dst = NULL;
1168 	}
1169 
1170 	if (err > 0)
1171 		err = np->recverr ? net_xmit_errno(err) : 0;
1172 	release_sock(sk);
1173 out:
1174 	dst_release(dst);
1175 	fl6_sock_release(flowlabel);
1176 	if (!err)
1177 		return len;
1178 	/*
1179 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1180 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1181 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1182 	 * things).  We could add another new stat but at least for now that
1183 	 * seems like overkill.
1184 	 */
1185 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1186 		UDP6_INC_STATS_USER(sock_net(sk),
1187 				UDP_MIB_SNDBUFERRORS, is_udplite);
1188 	}
1189 	return err;
1190 
1191 do_confirm:
1192 	dst_confirm(dst);
1193 	if (!(msg->msg_flags&MSG_PROBE) || len)
1194 		goto back_from_confirm;
1195 	err = 0;
1196 	goto out;
1197 }
1198 
1199 void udpv6_destroy_sock(struct sock *sk)
1200 {
1201 	lock_sock(sk);
1202 	udp_v6_flush_pending_frames(sk);
1203 	release_sock(sk);
1204 
1205 	inet6_destroy_sock(sk);
1206 }
1207 
1208 /*
1209  *	Socket option code for UDP
1210  */
1211 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1212 		     char __user *optval, unsigned int optlen)
1213 {
1214 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1215 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1216 					  udp_v6_push_pending_frames);
1217 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
1218 }
1219 
1220 #ifdef CONFIG_COMPAT
1221 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1222 			    char __user *optval, unsigned int optlen)
1223 {
1224 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1225 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1226 					  udp_v6_push_pending_frames);
1227 	return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1228 }
1229 #endif
1230 
1231 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1232 		     char __user *optval, int __user *optlen)
1233 {
1234 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1235 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1236 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
1237 }
1238 
1239 #ifdef CONFIG_COMPAT
1240 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1241 			    char __user *optval, int __user *optlen)
1242 {
1243 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1244 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1245 	return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1246 }
1247 #endif
1248 
1249 static int udp6_ufo_send_check(struct sk_buff *skb)
1250 {
1251 	struct ipv6hdr *ipv6h;
1252 	struct udphdr *uh;
1253 
1254 	if (!pskb_may_pull(skb, sizeof(*uh)))
1255 		return -EINVAL;
1256 
1257 	ipv6h = ipv6_hdr(skb);
1258 	uh = udp_hdr(skb);
1259 
1260 	uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
1261 				     IPPROTO_UDP, 0);
1262 	skb->csum_start = skb_transport_header(skb) - skb->head;
1263 	skb->csum_offset = offsetof(struct udphdr, check);
1264 	skb->ip_summed = CHECKSUM_PARTIAL;
1265 	return 0;
1266 }
1267 
1268 static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb, int features)
1269 {
1270 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1271 	unsigned int mss;
1272 	unsigned int unfrag_ip6hlen, unfrag_len;
1273 	struct frag_hdr *fptr;
1274 	u8 *mac_start, *prevhdr;
1275 	u8 nexthdr;
1276 	u8 frag_hdr_sz = sizeof(struct frag_hdr);
1277 	int offset;
1278 	__wsum csum;
1279 
1280 	mss = skb_shinfo(skb)->gso_size;
1281 	if (unlikely(skb->len <= mss))
1282 		goto out;
1283 
1284 	if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1285 		/* Packet is from an untrusted source, reset gso_segs. */
1286 		int type = skb_shinfo(skb)->gso_type;
1287 
1288 		if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1289 			     !(type & (SKB_GSO_UDP))))
1290 			goto out;
1291 
1292 		skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1293 
1294 		segs = NULL;
1295 		goto out;
1296 	}
1297 
1298 	/* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1299 	 * do checksum of UDP packets sent as multiple IP fragments.
1300 	 */
1301 	offset = skb->csum_start - skb_headroom(skb);
1302 	csum = skb_checksum(skb, offset, skb->len- offset, 0);
1303 	offset += skb->csum_offset;
1304 	*(__sum16 *)(skb->data + offset) = csum_fold(csum);
1305 	skb->ip_summed = CHECKSUM_NONE;
1306 
1307 	/* Check if there is enough headroom to insert fragment header. */
1308 	if ((skb_headroom(skb) < frag_hdr_sz) &&
1309 	    pskb_expand_head(skb, frag_hdr_sz, 0, GFP_ATOMIC))
1310 		goto out;
1311 
1312 	/* Find the unfragmentable header and shift it left by frag_hdr_sz
1313 	 * bytes to insert fragment header.
1314 	 */
1315 	unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
1316 	nexthdr = *prevhdr;
1317 	*prevhdr = NEXTHDR_FRAGMENT;
1318 	unfrag_len = skb_network_header(skb) - skb_mac_header(skb) +
1319 		     unfrag_ip6hlen;
1320 	mac_start = skb_mac_header(skb);
1321 	memmove(mac_start-frag_hdr_sz, mac_start, unfrag_len);
1322 
1323 	skb->mac_header -= frag_hdr_sz;
1324 	skb->network_header -= frag_hdr_sz;
1325 
1326 	fptr = (struct frag_hdr *)(skb_network_header(skb) + unfrag_ip6hlen);
1327 	fptr->nexthdr = nexthdr;
1328 	fptr->reserved = 0;
1329 	ipv6_select_ident(fptr);
1330 
1331 	/* Fragment the skb. ipv6 header and the remaining fields of the
1332 	 * fragment header are updated in ipv6_gso_segment()
1333 	 */
1334 	segs = skb_segment(skb, features);
1335 
1336 out:
1337 	return segs;
1338 }
1339 
1340 static const struct inet6_protocol udpv6_protocol = {
1341 	.handler	=	udpv6_rcv,
1342 	.err_handler	=	udpv6_err,
1343 	.gso_send_check =	udp6_ufo_send_check,
1344 	.gso_segment	=	udp6_ufo_fragment,
1345 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1346 };
1347 
1348 /* ------------------------------------------------------------------------ */
1349 #ifdef CONFIG_PROC_FS
1350 
1351 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
1352 {
1353 	struct inet_sock *inet = inet_sk(sp);
1354 	struct ipv6_pinfo *np = inet6_sk(sp);
1355 	struct in6_addr *dest, *src;
1356 	__u16 destp, srcp;
1357 
1358 	dest  = &np->daddr;
1359 	src   = &np->rcv_saddr;
1360 	destp = ntohs(inet->inet_dport);
1361 	srcp  = ntohs(inet->inet_sport);
1362 	seq_printf(seq,
1363 		   "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1364 		   "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d\n",
1365 		   bucket,
1366 		   src->s6_addr32[0], src->s6_addr32[1],
1367 		   src->s6_addr32[2], src->s6_addr32[3], srcp,
1368 		   dest->s6_addr32[0], dest->s6_addr32[1],
1369 		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
1370 		   sp->sk_state,
1371 		   sk_wmem_alloc_get(sp),
1372 		   sk_rmem_alloc_get(sp),
1373 		   0, 0L, 0,
1374 		   sock_i_uid(sp), 0,
1375 		   sock_i_ino(sp),
1376 		   atomic_read(&sp->sk_refcnt), sp,
1377 		   atomic_read(&sp->sk_drops));
1378 }
1379 
1380 int udp6_seq_show(struct seq_file *seq, void *v)
1381 {
1382 	if (v == SEQ_START_TOKEN)
1383 		seq_printf(seq,
1384 			   "  sl  "
1385 			   "local_address                         "
1386 			   "remote_address                        "
1387 			   "st tx_queue rx_queue tr tm->when retrnsmt"
1388 			   "   uid  timeout inode ref pointer drops\n");
1389 	else
1390 		udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1391 	return 0;
1392 }
1393 
1394 static struct udp_seq_afinfo udp6_seq_afinfo = {
1395 	.name		= "udp6",
1396 	.family		= AF_INET6,
1397 	.udp_table	= &udp_table,
1398 	.seq_fops	= {
1399 		.owner	=	THIS_MODULE,
1400 	},
1401 	.seq_ops	= {
1402 		.show		= udp6_seq_show,
1403 	},
1404 };
1405 
1406 int __net_init udp6_proc_init(struct net *net)
1407 {
1408 	return udp_proc_register(net, &udp6_seq_afinfo);
1409 }
1410 
1411 void udp6_proc_exit(struct net *net) {
1412 	udp_proc_unregister(net, &udp6_seq_afinfo);
1413 }
1414 #endif /* CONFIG_PROC_FS */
1415 
1416 /* ------------------------------------------------------------------------ */
1417 
1418 struct proto udpv6_prot = {
1419 	.name		   = "UDPv6",
1420 	.owner		   = THIS_MODULE,
1421 	.close		   = udp_lib_close,
1422 	.connect	   = ip6_datagram_connect,
1423 	.disconnect	   = udp_disconnect,
1424 	.ioctl		   = udp_ioctl,
1425 	.destroy	   = udpv6_destroy_sock,
1426 	.setsockopt	   = udpv6_setsockopt,
1427 	.getsockopt	   = udpv6_getsockopt,
1428 	.sendmsg	   = udpv6_sendmsg,
1429 	.recvmsg	   = udpv6_recvmsg,
1430 	.backlog_rcv	   = udpv6_queue_rcv_skb,
1431 	.hash		   = udp_lib_hash,
1432 	.unhash		   = udp_lib_unhash,
1433 	.get_port	   = udp_v6_get_port,
1434 	.memory_allocated  = &udp_memory_allocated,
1435 	.sysctl_mem	   = sysctl_udp_mem,
1436 	.sysctl_wmem	   = &sysctl_udp_wmem_min,
1437 	.sysctl_rmem	   = &sysctl_udp_rmem_min,
1438 	.obj_size	   = sizeof(struct udp6_sock),
1439 	.slab_flags	   = SLAB_DESTROY_BY_RCU,
1440 	.h.udp_table	   = &udp_table,
1441 #ifdef CONFIG_COMPAT
1442 	.compat_setsockopt = compat_udpv6_setsockopt,
1443 	.compat_getsockopt = compat_udpv6_getsockopt,
1444 #endif
1445 };
1446 
1447 static struct inet_protosw udpv6_protosw = {
1448 	.type =      SOCK_DGRAM,
1449 	.protocol =  IPPROTO_UDP,
1450 	.prot =      &udpv6_prot,
1451 	.ops =       &inet6_dgram_ops,
1452 	.no_check =  UDP_CSUM_DEFAULT,
1453 	.flags =     INET_PROTOSW_PERMANENT,
1454 };
1455 
1456 
1457 int __init udpv6_init(void)
1458 {
1459 	int ret;
1460 
1461 	ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1462 	if (ret)
1463 		goto out;
1464 
1465 	ret = inet6_register_protosw(&udpv6_protosw);
1466 	if (ret)
1467 		goto out_udpv6_protocol;
1468 out:
1469 	return ret;
1470 
1471 out_udpv6_protocol:
1472 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1473 	goto out;
1474 }
1475 
1476 void udpv6_exit(void)
1477 {
1478 	inet6_unregister_protosw(&udpv6_protosw);
1479 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1480 }
1481