xref: /openbmc/linux/net/ipv6/udp.c (revision d5532ee7)
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((__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 				struct in6_addr *saddr, __be16 sport,
126 				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(&result->sk_refcnt)))
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 				      struct in6_addr *saddr, __be16 sport,
243 				      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(&result->sk_refcnt)))
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 	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 /*
324  * 	This should be easy, if there is something there we
325  * 	return it, otherwise we block.
326  */
327 
328 int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk,
329 		  struct msghdr *msg, size_t len,
330 		  int noblock, int flags, int *addr_len)
331 {
332 	struct ipv6_pinfo *np = inet6_sk(sk);
333 	struct inet_sock *inet = inet_sk(sk);
334 	struct sk_buff *skb;
335 	unsigned int ulen;
336 	int peeked;
337 	int err;
338 	int is_udplite = IS_UDPLITE(sk);
339 	int is_udp4;
340 	bool slow;
341 
342 	if (addr_len)
343 		*addr_len=sizeof(struct sockaddr_in6);
344 
345 	if (flags & MSG_ERRQUEUE)
346 		return ipv6_recv_error(sk, msg, len);
347 
348 	if (np->rxpmtu && np->rxopt.bits.rxpmtu)
349 		return ipv6_recv_rxpmtu(sk, msg, len);
350 
351 try_again:
352 	skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
353 				  &peeked, &err);
354 	if (!skb)
355 		goto out;
356 
357 	ulen = skb->len - sizeof(struct udphdr);
358 	if (len > ulen)
359 		len = ulen;
360 	else if (len < ulen)
361 		msg->msg_flags |= MSG_TRUNC;
362 
363 	is_udp4 = (skb->protocol == htons(ETH_P_IP));
364 
365 	/*
366 	 * If checksum is needed at all, try to do it while copying the
367 	 * data.  If the data is truncated, or if we only want a partial
368 	 * coverage checksum (UDP-Lite), do it before the copy.
369 	 */
370 
371 	if (len < ulen || UDP_SKB_CB(skb)->partial_cov) {
372 		if (udp_lib_checksum_complete(skb))
373 			goto csum_copy_err;
374 	}
375 
376 	if (skb_csum_unnecessary(skb))
377 		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
378 					      msg->msg_iov,len);
379 	else {
380 		err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
381 		if (err == -EINVAL)
382 			goto csum_copy_err;
383 	}
384 	if (err)
385 		goto out_free;
386 
387 	if (!peeked) {
388 		if (is_udp4)
389 			UDP_INC_STATS_USER(sock_net(sk),
390 					UDP_MIB_INDATAGRAMS, is_udplite);
391 		else
392 			UDP6_INC_STATS_USER(sock_net(sk),
393 					UDP_MIB_INDATAGRAMS, is_udplite);
394 	}
395 
396 	sock_recv_ts_and_drops(msg, sk, skb);
397 
398 	/* Copy the address. */
399 	if (msg->msg_name) {
400 		struct sockaddr_in6 *sin6;
401 
402 		sin6 = (struct sockaddr_in6 *) msg->msg_name;
403 		sin6->sin6_family = AF_INET6;
404 		sin6->sin6_port = udp_hdr(skb)->source;
405 		sin6->sin6_flowinfo = 0;
406 		sin6->sin6_scope_id = 0;
407 
408 		if (is_udp4)
409 			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
410 					       &sin6->sin6_addr);
411 		else {
412 			ipv6_addr_copy(&sin6->sin6_addr,
413 				       &ipv6_hdr(skb)->saddr);
414 			if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
415 				sin6->sin6_scope_id = IP6CB(skb)->iif;
416 		}
417 
418 	}
419 	if (is_udp4) {
420 		if (inet->cmsg_flags)
421 			ip_cmsg_recv(msg, skb);
422 	} else {
423 		if (np->rxopt.all)
424 			datagram_recv_ctl(sk, msg, skb);
425 	}
426 
427 	err = len;
428 	if (flags & MSG_TRUNC)
429 		err = ulen;
430 
431 out_free:
432 	skb_free_datagram_locked(sk, skb);
433 out:
434 	return err;
435 
436 csum_copy_err:
437 	slow = lock_sock_fast(sk);
438 	if (!skb_kill_datagram(sk, skb, flags)) {
439 		if (is_udp4)
440 			UDP_INC_STATS_USER(sock_net(sk),
441 					UDP_MIB_INERRORS, is_udplite);
442 		else
443 			UDP6_INC_STATS_USER(sock_net(sk),
444 					UDP_MIB_INERRORS, is_udplite);
445 	}
446 	unlock_sock_fast(sk, slow);
447 
448 	if (flags & MSG_DONTWAIT)
449 		return -EAGAIN;
450 	goto try_again;
451 }
452 
453 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
454 		    u8 type, u8 code, int offset, __be32 info,
455 		    struct udp_table *udptable)
456 {
457 	struct ipv6_pinfo *np;
458 	struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
459 	struct in6_addr *saddr = &hdr->saddr;
460 	struct in6_addr *daddr = &hdr->daddr;
461 	struct udphdr *uh = (struct udphdr*)(skb->data+offset);
462 	struct sock *sk;
463 	int err;
464 
465 	sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest,
466 			       saddr, uh->source, inet6_iif(skb), udptable);
467 	if (sk == NULL)
468 		return;
469 
470 	np = inet6_sk(sk);
471 
472 	if (!icmpv6_err_convert(type, code, &err) && !np->recverr)
473 		goto out;
474 
475 	if (sk->sk_state != TCP_ESTABLISHED && !np->recverr)
476 		goto out;
477 
478 	if (np->recverr)
479 		ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
480 
481 	sk->sk_err = err;
482 	sk->sk_error_report(sk);
483 out:
484 	sock_put(sk);
485 }
486 
487 static __inline__ void udpv6_err(struct sk_buff *skb,
488 				 struct inet6_skb_parm *opt, u8 type,
489 				 u8 code, int offset, __be32 info     )
490 {
491 	__udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
492 }
493 
494 int udpv6_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
495 {
496 	struct udp_sock *up = udp_sk(sk);
497 	int rc;
498 	int is_udplite = IS_UDPLITE(sk);
499 
500 	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
501 		goto drop;
502 
503 	/*
504 	 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
505 	 */
506 	if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
507 
508 		if (up->pcrlen == 0) {          /* full coverage was set  */
509 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage"
510 				" %d while full coverage %d requested\n",
511 				UDP_SKB_CB(skb)->cscov, skb->len);
512 			goto drop;
513 		}
514 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
515 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d "
516 						    "too small, need min %d\n",
517 				       UDP_SKB_CB(skb)->cscov, up->pcrlen);
518 			goto drop;
519 		}
520 	}
521 
522 	if (sk->sk_filter) {
523 		if (udp_lib_checksum_complete(skb))
524 			goto drop;
525 	}
526 
527 	if ((rc = ip_queue_rcv_skb(sk, skb)) < 0) {
528 		/* Note that an ENOMEM error is charged twice */
529 		if (rc == -ENOMEM)
530 			UDP6_INC_STATS_BH(sock_net(sk),
531 					UDP_MIB_RCVBUFERRORS, is_udplite);
532 		goto drop_no_sk_drops_inc;
533 	}
534 
535 	return 0;
536 drop:
537 	atomic_inc(&sk->sk_drops);
538 drop_no_sk_drops_inc:
539 	UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
540 	kfree_skb(skb);
541 	return -1;
542 }
543 
544 static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk,
545 				      __be16 loc_port, struct in6_addr *loc_addr,
546 				      __be16 rmt_port, struct in6_addr *rmt_addr,
547 				      int dif)
548 {
549 	struct hlist_nulls_node *node;
550 	struct sock *s = sk;
551 	unsigned short num = ntohs(loc_port);
552 
553 	sk_nulls_for_each_from(s, node) {
554 		struct inet_sock *inet = inet_sk(s);
555 
556 		if (!net_eq(sock_net(s), net))
557 			continue;
558 
559 		if (udp_sk(s)->udp_port_hash == num &&
560 		    s->sk_family == PF_INET6) {
561 			struct ipv6_pinfo *np = inet6_sk(s);
562 			if (inet->inet_dport) {
563 				if (inet->inet_dport != rmt_port)
564 					continue;
565 			}
566 			if (!ipv6_addr_any(&np->daddr) &&
567 			    !ipv6_addr_equal(&np->daddr, rmt_addr))
568 				continue;
569 
570 			if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)
571 				continue;
572 
573 			if (!ipv6_addr_any(&np->rcv_saddr)) {
574 				if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr))
575 					continue;
576 			}
577 			if (!inet6_mc_check(s, loc_addr, rmt_addr))
578 				continue;
579 			return s;
580 		}
581 	}
582 	return NULL;
583 }
584 
585 static void flush_stack(struct sock **stack, unsigned int count,
586 			struct sk_buff *skb, unsigned int final)
587 {
588 	unsigned int i;
589 	struct sock *sk;
590 	struct sk_buff *skb1;
591 
592 	for (i = 0; i < count; i++) {
593 		skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
594 
595 		sk = stack[i];
596 		if (skb1) {
597 			if (sk_rcvqueues_full(sk, skb)) {
598 				kfree_skb(skb1);
599 				goto drop;
600 			}
601 			bh_lock_sock(sk);
602 			if (!sock_owned_by_user(sk))
603 				udpv6_queue_rcv_skb(sk, skb1);
604 			else if (sk_add_backlog(sk, skb1)) {
605 				kfree_skb(skb1);
606 				bh_unlock_sock(sk);
607 				goto drop;
608 			}
609 			bh_unlock_sock(sk);
610 			continue;
611 		}
612 drop:
613 		atomic_inc(&sk->sk_drops);
614 		UDP6_INC_STATS_BH(sock_net(sk),
615 				UDP_MIB_RCVBUFERRORS, IS_UDPLITE(sk));
616 		UDP6_INC_STATS_BH(sock_net(sk),
617 				UDP_MIB_INERRORS, IS_UDPLITE(sk));
618 	}
619 }
620 /*
621  * Note: called only from the BH handler context,
622  * so we don't need to lock the hashes.
623  */
624 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
625 		struct in6_addr *saddr, struct in6_addr *daddr,
626 		struct udp_table *udptable)
627 {
628 	struct sock *sk, *stack[256 / sizeof(struct sock *)];
629 	const struct udphdr *uh = udp_hdr(skb);
630 	struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
631 	int dif;
632 	unsigned int i, count = 0;
633 
634 	spin_lock(&hslot->lock);
635 	sk = sk_nulls_head(&hslot->head);
636 	dif = inet6_iif(skb);
637 	sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
638 	while (sk) {
639 		stack[count++] = sk;
640 		sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr,
641 				       uh->source, saddr, dif);
642 		if (unlikely(count == ARRAY_SIZE(stack))) {
643 			if (!sk)
644 				break;
645 			flush_stack(stack, count, skb, ~0);
646 			count = 0;
647 		}
648 	}
649 	/*
650 	 * before releasing the lock, we must take reference on sockets
651 	 */
652 	for (i = 0; i < count; i++)
653 		sock_hold(stack[i]);
654 
655 	spin_unlock(&hslot->lock);
656 
657 	if (count) {
658 		flush_stack(stack, count, skb, count - 1);
659 
660 		for (i = 0; i < count; i++)
661 			sock_put(stack[i]);
662 	} else {
663 		kfree_skb(skb);
664 	}
665 	return 0;
666 }
667 
668 static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh,
669 				 int proto)
670 {
671 	int err;
672 
673 	UDP_SKB_CB(skb)->partial_cov = 0;
674 	UDP_SKB_CB(skb)->cscov = skb->len;
675 
676 	if (proto == IPPROTO_UDPLITE) {
677 		err = udplite_checksum_init(skb, uh);
678 		if (err)
679 			return err;
680 	}
681 
682 	if (uh->check == 0) {
683 		/* RFC 2460 section 8.1 says that we SHOULD log
684 		   this error. Well, it is reasonable.
685 		 */
686 		LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n");
687 		return 1;
688 	}
689 	if (skb->ip_summed == CHECKSUM_COMPLETE &&
690 	    !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
691 			     skb->len, proto, skb->csum))
692 		skb->ip_summed = CHECKSUM_UNNECESSARY;
693 
694 	if (!skb_csum_unnecessary(skb))
695 		skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
696 							 &ipv6_hdr(skb)->daddr,
697 							 skb->len, proto, 0));
698 
699 	return 0;
700 }
701 
702 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
703 		   int proto)
704 {
705 	struct net *net = dev_net(skb->dev);
706 	struct sock *sk;
707 	struct udphdr *uh;
708 	struct in6_addr *saddr, *daddr;
709 	u32 ulen = 0;
710 
711 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
712 		goto discard;
713 
714 	saddr = &ipv6_hdr(skb)->saddr;
715 	daddr = &ipv6_hdr(skb)->daddr;
716 	uh = udp_hdr(skb);
717 
718 	ulen = ntohs(uh->len);
719 	if (ulen > skb->len)
720 		goto short_packet;
721 
722 	if (proto == IPPROTO_UDP) {
723 		/* UDP validates ulen. */
724 
725 		/* Check for jumbo payload */
726 		if (ulen == 0)
727 			ulen = skb->len;
728 
729 		if (ulen < sizeof(*uh))
730 			goto short_packet;
731 
732 		if (ulen < skb->len) {
733 			if (pskb_trim_rcsum(skb, ulen))
734 				goto short_packet;
735 			saddr = &ipv6_hdr(skb)->saddr;
736 			daddr = &ipv6_hdr(skb)->daddr;
737 			uh = udp_hdr(skb);
738 		}
739 	}
740 
741 	if (udp6_csum_init(skb, uh, proto))
742 		goto discard;
743 
744 	/*
745 	 *	Multicast receive code
746 	 */
747 	if (ipv6_addr_is_multicast(daddr))
748 		return __udp6_lib_mcast_deliver(net, skb,
749 				saddr, daddr, udptable);
750 
751 	/* Unicast */
752 
753 	/*
754 	 * check socket cache ... must talk to Alan about his plans
755 	 * for sock caches... i'll skip this for now.
756 	 */
757 	sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
758 
759 	if (sk == NULL) {
760 		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
761 			goto discard;
762 
763 		if (udp_lib_checksum_complete(skb))
764 			goto discard;
765 		UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS,
766 				proto == IPPROTO_UDPLITE);
767 
768 		icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
769 
770 		kfree_skb(skb);
771 		return 0;
772 	}
773 
774 	/* deliver */
775 
776 	if (sk_rcvqueues_full(sk, skb)) {
777 		sock_put(sk);
778 		goto discard;
779 	}
780 	bh_lock_sock(sk);
781 	if (!sock_owned_by_user(sk))
782 		udpv6_queue_rcv_skb(sk, skb);
783 	else if (sk_add_backlog(sk, skb)) {
784 		atomic_inc(&sk->sk_drops);
785 		bh_unlock_sock(sk);
786 		sock_put(sk);
787 		goto discard;
788 	}
789 	bh_unlock_sock(sk);
790 	sock_put(sk);
791 	return 0;
792 
793 short_packet:
794 	LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
795 		       proto == IPPROTO_UDPLITE ? "-Lite" : "",
796 		       saddr,
797 		       ntohs(uh->source),
798 		       ulen,
799 		       skb->len,
800 		       daddr,
801 		       ntohs(uh->dest));
802 
803 discard:
804 	UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
805 	kfree_skb(skb);
806 	return 0;
807 }
808 
809 static __inline__ int udpv6_rcv(struct sk_buff *skb)
810 {
811 	return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
812 }
813 
814 /*
815  * Throw away all pending data and cancel the corking. Socket is locked.
816  */
817 static void udp_v6_flush_pending_frames(struct sock *sk)
818 {
819 	struct udp_sock *up = udp_sk(sk);
820 
821 	if (up->pending == AF_INET)
822 		udp_flush_pending_frames(sk);
823 	else if (up->pending) {
824 		up->len = 0;
825 		up->pending = 0;
826 		ip6_flush_pending_frames(sk);
827 	}
828 }
829 
830 /**
831  * 	udp6_hwcsum_outgoing  -  handle outgoing HW checksumming
832  * 	@sk: 	socket we are sending on
833  * 	@skb: 	sk_buff containing the filled-in UDP header
834  * 	        (checksum field must be zeroed out)
835  */
836 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
837 				 const struct in6_addr *saddr,
838 				 const struct in6_addr *daddr, int len)
839 {
840 	unsigned int offset;
841 	struct udphdr *uh = udp_hdr(skb);
842 	__wsum csum = 0;
843 
844 	if (skb_queue_len(&sk->sk_write_queue) == 1) {
845 		/* Only one fragment on the socket.  */
846 		skb->csum_start = skb_transport_header(skb) - skb->head;
847 		skb->csum_offset = offsetof(struct udphdr, check);
848 		uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
849 	} else {
850 		/*
851 		 * HW-checksum won't work as there are two or more
852 		 * fragments on the socket so that all csums of sk_buffs
853 		 * should be together
854 		 */
855 		offset = skb_transport_offset(skb);
856 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
857 
858 		skb->ip_summed = CHECKSUM_NONE;
859 
860 		skb_queue_walk(&sk->sk_write_queue, skb) {
861 			csum = csum_add(csum, skb->csum);
862 		}
863 
864 		uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
865 					    csum);
866 		if (uh->check == 0)
867 			uh->check = CSUM_MANGLED_0;
868 	}
869 }
870 
871 /*
872  *	Sending
873  */
874 
875 static int udp_v6_push_pending_frames(struct sock *sk)
876 {
877 	struct sk_buff *skb;
878 	struct udphdr *uh;
879 	struct udp_sock  *up = udp_sk(sk);
880 	struct inet_sock *inet = inet_sk(sk);
881 	struct flowi *fl = &inet->cork.fl;
882 	int err = 0;
883 	int is_udplite = IS_UDPLITE(sk);
884 	__wsum csum = 0;
885 
886 	/* Grab the skbuff where UDP header space exists. */
887 	if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
888 		goto out;
889 
890 	/*
891 	 * Create a UDP header
892 	 */
893 	uh = udp_hdr(skb);
894 	uh->source = fl->fl_ip_sport;
895 	uh->dest = fl->fl_ip_dport;
896 	uh->len = htons(up->len);
897 	uh->check = 0;
898 
899 	if (is_udplite)
900 		csum = udplite_csum_outgoing(sk, skb);
901 	else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
902 		udp6_hwcsum_outgoing(sk, skb, &fl->fl6_src, &fl->fl6_dst,
903 				     up->len);
904 		goto send;
905 	} else
906 		csum = udp_csum_outgoing(sk, skb);
907 
908 	/* add protocol-dependent pseudo-header */
909 	uh->check = csum_ipv6_magic(&fl->fl6_src, &fl->fl6_dst,
910 				    up->len, fl->proto, csum   );
911 	if (uh->check == 0)
912 		uh->check = CSUM_MANGLED_0;
913 
914 send:
915 	err = ip6_push_pending_frames(sk);
916 	if (err) {
917 		if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
918 			UDP6_INC_STATS_USER(sock_net(sk),
919 					    UDP_MIB_SNDBUFERRORS, is_udplite);
920 			err = 0;
921 		}
922 	} else
923 		UDP6_INC_STATS_USER(sock_net(sk),
924 				    UDP_MIB_OUTDATAGRAMS, is_udplite);
925 out:
926 	up->len = 0;
927 	up->pending = 0;
928 	return err;
929 }
930 
931 int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk,
932 		  struct msghdr *msg, size_t len)
933 {
934 	struct ipv6_txoptions opt_space;
935 	struct udp_sock *up = udp_sk(sk);
936 	struct inet_sock *inet = inet_sk(sk);
937 	struct ipv6_pinfo *np = inet6_sk(sk);
938 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name;
939 	struct in6_addr *daddr, *final_p, final;
940 	struct ipv6_txoptions *opt = NULL;
941 	struct ip6_flowlabel *flowlabel = NULL;
942 	struct flowi fl;
943 	struct dst_entry *dst;
944 	int addr_len = msg->msg_namelen;
945 	int ulen = len;
946 	int hlimit = -1;
947 	int tclass = -1;
948 	int dontfrag = -1;
949 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
950 	int err;
951 	int connected = 0;
952 	int is_udplite = IS_UDPLITE(sk);
953 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
954 
955 	/* destination address check */
956 	if (sin6) {
957 		if (addr_len < offsetof(struct sockaddr, sa_data))
958 			return -EINVAL;
959 
960 		switch (sin6->sin6_family) {
961 		case AF_INET6:
962 			if (addr_len < SIN6_LEN_RFC2133)
963 				return -EINVAL;
964 			daddr = &sin6->sin6_addr;
965 			break;
966 		case AF_INET:
967 			goto do_udp_sendmsg;
968 		case AF_UNSPEC:
969 			msg->msg_name = sin6 = NULL;
970 			msg->msg_namelen = addr_len = 0;
971 			daddr = NULL;
972 			break;
973 		default:
974 			return -EINVAL;
975 		}
976 	} else if (!up->pending) {
977 		if (sk->sk_state != TCP_ESTABLISHED)
978 			return -EDESTADDRREQ;
979 		daddr = &np->daddr;
980 	} else
981 		daddr = NULL;
982 
983 	if (daddr) {
984 		if (ipv6_addr_v4mapped(daddr)) {
985 			struct sockaddr_in sin;
986 			sin.sin_family = AF_INET;
987 			sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
988 			sin.sin_addr.s_addr = daddr->s6_addr32[3];
989 			msg->msg_name = &sin;
990 			msg->msg_namelen = sizeof(sin);
991 do_udp_sendmsg:
992 			if (__ipv6_only_sock(sk))
993 				return -ENETUNREACH;
994 			return udp_sendmsg(iocb, sk, msg, len);
995 		}
996 	}
997 
998 	if (up->pending == AF_INET)
999 		return udp_sendmsg(iocb, sk, msg, len);
1000 
1001 	/* Rough check on arithmetic overflow,
1002 	   better check is made in ip6_append_data().
1003 	   */
1004 	if (len > INT_MAX - sizeof(struct udphdr))
1005 		return -EMSGSIZE;
1006 
1007 	if (up->pending) {
1008 		/*
1009 		 * There are pending frames.
1010 		 * The socket lock must be held while it's corked.
1011 		 */
1012 		lock_sock(sk);
1013 		if (likely(up->pending)) {
1014 			if (unlikely(up->pending != AF_INET6)) {
1015 				release_sock(sk);
1016 				return -EAFNOSUPPORT;
1017 			}
1018 			dst = NULL;
1019 			goto do_append_data;
1020 		}
1021 		release_sock(sk);
1022 	}
1023 	ulen += sizeof(struct udphdr);
1024 
1025 	memset(&fl, 0, sizeof(fl));
1026 
1027 	if (sin6) {
1028 		if (sin6->sin6_port == 0)
1029 			return -EINVAL;
1030 
1031 		fl.fl_ip_dport = sin6->sin6_port;
1032 		daddr = &sin6->sin6_addr;
1033 
1034 		if (np->sndflow) {
1035 			fl.fl6_flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1036 			if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
1037 				flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
1038 				if (flowlabel == NULL)
1039 					return -EINVAL;
1040 				daddr = &flowlabel->dst;
1041 			}
1042 		}
1043 
1044 		/*
1045 		 * Otherwise it will be difficult to maintain
1046 		 * sk->sk_dst_cache.
1047 		 */
1048 		if (sk->sk_state == TCP_ESTABLISHED &&
1049 		    ipv6_addr_equal(daddr, &np->daddr))
1050 			daddr = &np->daddr;
1051 
1052 		if (addr_len >= sizeof(struct sockaddr_in6) &&
1053 		    sin6->sin6_scope_id &&
1054 		    ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL)
1055 			fl.oif = sin6->sin6_scope_id;
1056 	} else {
1057 		if (sk->sk_state != TCP_ESTABLISHED)
1058 			return -EDESTADDRREQ;
1059 
1060 		fl.fl_ip_dport = inet->inet_dport;
1061 		daddr = &np->daddr;
1062 		fl.fl6_flowlabel = np->flow_label;
1063 		connected = 1;
1064 	}
1065 
1066 	if (!fl.oif)
1067 		fl.oif = sk->sk_bound_dev_if;
1068 
1069 	if (!fl.oif)
1070 		fl.oif = np->sticky_pktinfo.ipi6_ifindex;
1071 
1072 	fl.mark = sk->sk_mark;
1073 
1074 	if (msg->msg_controllen) {
1075 		opt = &opt_space;
1076 		memset(opt, 0, sizeof(struct ipv6_txoptions));
1077 		opt->tot_len = sizeof(*opt);
1078 
1079 		err = datagram_send_ctl(sock_net(sk), msg, &fl, opt, &hlimit,
1080 					&tclass, &dontfrag);
1081 		if (err < 0) {
1082 			fl6_sock_release(flowlabel);
1083 			return err;
1084 		}
1085 		if ((fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1086 			flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
1087 			if (flowlabel == NULL)
1088 				return -EINVAL;
1089 		}
1090 		if (!(opt->opt_nflen|opt->opt_flen))
1091 			opt = NULL;
1092 		connected = 0;
1093 	}
1094 	if (opt == NULL)
1095 		opt = np->opt;
1096 	if (flowlabel)
1097 		opt = fl6_merge_options(&opt_space, flowlabel, opt);
1098 	opt = ipv6_fixup_options(&opt_space, opt);
1099 
1100 	fl.proto = sk->sk_protocol;
1101 	if (!ipv6_addr_any(daddr))
1102 		ipv6_addr_copy(&fl.fl6_dst, daddr);
1103 	else
1104 		fl.fl6_dst.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1105 	if (ipv6_addr_any(&fl.fl6_src) && !ipv6_addr_any(&np->saddr))
1106 		ipv6_addr_copy(&fl.fl6_src, &np->saddr);
1107 	fl.fl_ip_sport = inet->inet_sport;
1108 
1109 	final_p = fl6_update_dst(&fl, opt, &final);
1110 	if (final_p)
1111 		connected = 0;
1112 
1113 	if (!fl.oif && ipv6_addr_is_multicast(&fl.fl6_dst)) {
1114 		fl.oif = np->mcast_oif;
1115 		connected = 0;
1116 	}
1117 
1118 	security_sk_classify_flow(sk, &fl);
1119 
1120 	err = ip6_sk_dst_lookup(sk, &dst, &fl);
1121 	if (err)
1122 		goto out;
1123 	if (final_p)
1124 		ipv6_addr_copy(&fl.fl6_dst, final_p);
1125 
1126 	err = __xfrm_lookup(sock_net(sk), &dst, &fl, sk, XFRM_LOOKUP_WAIT);
1127 	if (err < 0) {
1128 		if (err == -EREMOTE)
1129 			err = ip6_dst_blackhole(sk, &dst, &fl);
1130 		if (err < 0)
1131 			goto out;
1132 	}
1133 
1134 	if (hlimit < 0) {
1135 		if (ipv6_addr_is_multicast(&fl.fl6_dst))
1136 			hlimit = np->mcast_hops;
1137 		else
1138 			hlimit = np->hop_limit;
1139 		if (hlimit < 0)
1140 			hlimit = ip6_dst_hoplimit(dst);
1141 	}
1142 
1143 	if (tclass < 0)
1144 		tclass = np->tclass;
1145 
1146 	if (dontfrag < 0)
1147 		dontfrag = np->dontfrag;
1148 
1149 	if (msg->msg_flags&MSG_CONFIRM)
1150 		goto do_confirm;
1151 back_from_confirm:
1152 
1153 	lock_sock(sk);
1154 	if (unlikely(up->pending)) {
1155 		/* The socket is already corked while preparing it. */
1156 		/* ... which is an evident application bug. --ANK */
1157 		release_sock(sk);
1158 
1159 		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
1160 		err = -EINVAL;
1161 		goto out;
1162 	}
1163 
1164 	up->pending = AF_INET6;
1165 
1166 do_append_data:
1167 	up->len += ulen;
1168 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
1169 	err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen,
1170 		sizeof(struct udphdr), hlimit, tclass, opt, &fl,
1171 		(struct rt6_info*)dst,
1172 		corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag);
1173 	if (err)
1174 		udp_v6_flush_pending_frames(sk);
1175 	else if (!corkreq)
1176 		err = udp_v6_push_pending_frames(sk);
1177 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1178 		up->pending = 0;
1179 
1180 	if (dst) {
1181 		if (connected) {
1182 			ip6_dst_store(sk, dst,
1183 				      ipv6_addr_equal(&fl.fl6_dst, &np->daddr) ?
1184 				      &np->daddr : NULL,
1185 #ifdef CONFIG_IPV6_SUBTREES
1186 				      ipv6_addr_equal(&fl.fl6_src, &np->saddr) ?
1187 				      &np->saddr :
1188 #endif
1189 				      NULL);
1190 		} else {
1191 			dst_release(dst);
1192 		}
1193 		dst = NULL;
1194 	}
1195 
1196 	if (err > 0)
1197 		err = np->recverr ? net_xmit_errno(err) : 0;
1198 	release_sock(sk);
1199 out:
1200 	dst_release(dst);
1201 	fl6_sock_release(flowlabel);
1202 	if (!err)
1203 		return len;
1204 	/*
1205 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1206 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1207 	 * we don't have a good statistic (IpOutDiscards but it can be too many
1208 	 * things).  We could add another new stat but at least for now that
1209 	 * seems like overkill.
1210 	 */
1211 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1212 		UDP6_INC_STATS_USER(sock_net(sk),
1213 				UDP_MIB_SNDBUFERRORS, is_udplite);
1214 	}
1215 	return err;
1216 
1217 do_confirm:
1218 	dst_confirm(dst);
1219 	if (!(msg->msg_flags&MSG_PROBE) || len)
1220 		goto back_from_confirm;
1221 	err = 0;
1222 	goto out;
1223 }
1224 
1225 void udpv6_destroy_sock(struct sock *sk)
1226 {
1227 	lock_sock(sk);
1228 	udp_v6_flush_pending_frames(sk);
1229 	release_sock(sk);
1230 
1231 	inet6_destroy_sock(sk);
1232 }
1233 
1234 /*
1235  *	Socket option code for UDP
1236  */
1237 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1238 		     char __user *optval, unsigned int optlen)
1239 {
1240 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1241 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1242 					  udp_v6_push_pending_frames);
1243 	return ipv6_setsockopt(sk, level, optname, optval, optlen);
1244 }
1245 
1246 #ifdef CONFIG_COMPAT
1247 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1248 			    char __user *optval, unsigned int optlen)
1249 {
1250 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1251 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1252 					  udp_v6_push_pending_frames);
1253 	return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1254 }
1255 #endif
1256 
1257 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1258 		     char __user *optval, int __user *optlen)
1259 {
1260 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1261 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1262 	return ipv6_getsockopt(sk, level, optname, optval, optlen);
1263 }
1264 
1265 #ifdef CONFIG_COMPAT
1266 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1267 			    char __user *optval, int __user *optlen)
1268 {
1269 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1270 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1271 	return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1272 }
1273 #endif
1274 
1275 static int udp6_ufo_send_check(struct sk_buff *skb)
1276 {
1277 	struct ipv6hdr *ipv6h;
1278 	struct udphdr *uh;
1279 
1280 	if (!pskb_may_pull(skb, sizeof(*uh)))
1281 		return -EINVAL;
1282 
1283 	ipv6h = ipv6_hdr(skb);
1284 	uh = udp_hdr(skb);
1285 
1286 	uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
1287 				     IPPROTO_UDP, 0);
1288 	skb->csum_start = skb_transport_header(skb) - skb->head;
1289 	skb->csum_offset = offsetof(struct udphdr, check);
1290 	skb->ip_summed = CHECKSUM_PARTIAL;
1291 	return 0;
1292 }
1293 
1294 static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb, int features)
1295 {
1296 	struct sk_buff *segs = ERR_PTR(-EINVAL);
1297 	unsigned int mss;
1298 	unsigned int unfrag_ip6hlen, unfrag_len;
1299 	struct frag_hdr *fptr;
1300 	u8 *mac_start, *prevhdr;
1301 	u8 nexthdr;
1302 	u8 frag_hdr_sz = sizeof(struct frag_hdr);
1303 	int offset;
1304 	__wsum csum;
1305 
1306 	mss = skb_shinfo(skb)->gso_size;
1307 	if (unlikely(skb->len <= mss))
1308 		goto out;
1309 
1310 	if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1311 		/* Packet is from an untrusted source, reset gso_segs. */
1312 		int type = skb_shinfo(skb)->gso_type;
1313 
1314 		if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1315 			     !(type & (SKB_GSO_UDP))))
1316 			goto out;
1317 
1318 		skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1319 
1320 		segs = NULL;
1321 		goto out;
1322 	}
1323 
1324 	/* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1325 	 * do checksum of UDP packets sent as multiple IP fragments.
1326 	 */
1327 	offset = skb->csum_start - skb_headroom(skb);
1328 	csum = skb_checksum(skb, offset, skb->len- offset, 0);
1329 	offset += skb->csum_offset;
1330 	*(__sum16 *)(skb->data + offset) = csum_fold(csum);
1331 	skb->ip_summed = CHECKSUM_NONE;
1332 
1333 	/* Check if there is enough headroom to insert fragment header. */
1334 	if ((skb_headroom(skb) < frag_hdr_sz) &&
1335 	    pskb_expand_head(skb, frag_hdr_sz, 0, GFP_ATOMIC))
1336 		goto out;
1337 
1338 	/* Find the unfragmentable header and shift it left by frag_hdr_sz
1339 	 * bytes to insert fragment header.
1340 	 */
1341 	unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
1342 	nexthdr = *prevhdr;
1343 	*prevhdr = NEXTHDR_FRAGMENT;
1344 	unfrag_len = skb_network_header(skb) - skb_mac_header(skb) +
1345 		     unfrag_ip6hlen;
1346 	mac_start = skb_mac_header(skb);
1347 	memmove(mac_start-frag_hdr_sz, mac_start, unfrag_len);
1348 
1349 	skb->mac_header -= frag_hdr_sz;
1350 	skb->network_header -= frag_hdr_sz;
1351 
1352 	fptr = (struct frag_hdr *)(skb_network_header(skb) + unfrag_ip6hlen);
1353 	fptr->nexthdr = nexthdr;
1354 	fptr->reserved = 0;
1355 	ipv6_select_ident(fptr);
1356 
1357 	/* Fragment the skb. ipv6 header and the remaining fields of the
1358 	 * fragment header are updated in ipv6_gso_segment()
1359 	 */
1360 	segs = skb_segment(skb, features);
1361 
1362 out:
1363 	return segs;
1364 }
1365 
1366 static const struct inet6_protocol udpv6_protocol = {
1367 	.handler	=	udpv6_rcv,
1368 	.err_handler	=	udpv6_err,
1369 	.gso_send_check =	udp6_ufo_send_check,
1370 	.gso_segment	=	udp6_ufo_fragment,
1371 	.flags		=	INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1372 };
1373 
1374 /* ------------------------------------------------------------------------ */
1375 #ifdef CONFIG_PROC_FS
1376 
1377 static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket)
1378 {
1379 	struct inet_sock *inet = inet_sk(sp);
1380 	struct ipv6_pinfo *np = inet6_sk(sp);
1381 	struct in6_addr *dest, *src;
1382 	__u16 destp, srcp;
1383 
1384 	dest  = &np->daddr;
1385 	src   = &np->rcv_saddr;
1386 	destp = ntohs(inet->inet_dport);
1387 	srcp  = ntohs(inet->inet_sport);
1388 	seq_printf(seq,
1389 		   "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1390 		   "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d\n",
1391 		   bucket,
1392 		   src->s6_addr32[0], src->s6_addr32[1],
1393 		   src->s6_addr32[2], src->s6_addr32[3], srcp,
1394 		   dest->s6_addr32[0], dest->s6_addr32[1],
1395 		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
1396 		   sp->sk_state,
1397 		   sk_wmem_alloc_get(sp),
1398 		   sk_rmem_alloc_get(sp),
1399 		   0, 0L, 0,
1400 		   sock_i_uid(sp), 0,
1401 		   sock_i_ino(sp),
1402 		   atomic_read(&sp->sk_refcnt), sp,
1403 		   atomic_read(&sp->sk_drops));
1404 }
1405 
1406 int udp6_seq_show(struct seq_file *seq, void *v)
1407 {
1408 	if (v == SEQ_START_TOKEN)
1409 		seq_printf(seq,
1410 			   "  sl  "
1411 			   "local_address                         "
1412 			   "remote_address                        "
1413 			   "st tx_queue rx_queue tr tm->when retrnsmt"
1414 			   "   uid  timeout inode ref pointer drops\n");
1415 	else
1416 		udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket);
1417 	return 0;
1418 }
1419 
1420 static struct udp_seq_afinfo udp6_seq_afinfo = {
1421 	.name		= "udp6",
1422 	.family		= AF_INET6,
1423 	.udp_table	= &udp_table,
1424 	.seq_fops	= {
1425 		.owner	=	THIS_MODULE,
1426 	},
1427 	.seq_ops	= {
1428 		.show		= udp6_seq_show,
1429 	},
1430 };
1431 
1432 int __net_init udp6_proc_init(struct net *net)
1433 {
1434 	return udp_proc_register(net, &udp6_seq_afinfo);
1435 }
1436 
1437 void udp6_proc_exit(struct net *net) {
1438 	udp_proc_unregister(net, &udp6_seq_afinfo);
1439 }
1440 #endif /* CONFIG_PROC_FS */
1441 
1442 /* ------------------------------------------------------------------------ */
1443 
1444 struct proto udpv6_prot = {
1445 	.name		   = "UDPv6",
1446 	.owner		   = THIS_MODULE,
1447 	.close		   = udp_lib_close,
1448 	.connect	   = ip6_datagram_connect,
1449 	.disconnect	   = udp_disconnect,
1450 	.ioctl		   = udp_ioctl,
1451 	.destroy	   = udpv6_destroy_sock,
1452 	.setsockopt	   = udpv6_setsockopt,
1453 	.getsockopt	   = udpv6_getsockopt,
1454 	.sendmsg	   = udpv6_sendmsg,
1455 	.recvmsg	   = udpv6_recvmsg,
1456 	.backlog_rcv	   = udpv6_queue_rcv_skb,
1457 	.hash		   = udp_lib_hash,
1458 	.unhash		   = udp_lib_unhash,
1459 	.rehash		   = udp_v6_rehash,
1460 	.get_port	   = udp_v6_get_port,
1461 	.memory_allocated  = &udp_memory_allocated,
1462 	.sysctl_mem	   = sysctl_udp_mem,
1463 	.sysctl_wmem	   = &sysctl_udp_wmem_min,
1464 	.sysctl_rmem	   = &sysctl_udp_rmem_min,
1465 	.obj_size	   = sizeof(struct udp6_sock),
1466 	.slab_flags	   = SLAB_DESTROY_BY_RCU,
1467 	.h.udp_table	   = &udp_table,
1468 #ifdef CONFIG_COMPAT
1469 	.compat_setsockopt = compat_udpv6_setsockopt,
1470 	.compat_getsockopt = compat_udpv6_getsockopt,
1471 #endif
1472 };
1473 
1474 static struct inet_protosw udpv6_protosw = {
1475 	.type =      SOCK_DGRAM,
1476 	.protocol =  IPPROTO_UDP,
1477 	.prot =      &udpv6_prot,
1478 	.ops =       &inet6_dgram_ops,
1479 	.no_check =  UDP_CSUM_DEFAULT,
1480 	.flags =     INET_PROTOSW_PERMANENT,
1481 };
1482 
1483 
1484 int __init udpv6_init(void)
1485 {
1486 	int ret;
1487 
1488 	ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1489 	if (ret)
1490 		goto out;
1491 
1492 	ret = inet6_register_protosw(&udpv6_protosw);
1493 	if (ret)
1494 		goto out_udpv6_protocol;
1495 out:
1496 	return ret;
1497 
1498 out_udpv6_protocol:
1499 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1500 	goto out;
1501 }
1502 
1503 void udpv6_exit(void)
1504 {
1505 	inet6_unregister_protosw(&udpv6_protosw);
1506 	inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1507 }
1508