xref: /linux/net/ipv4/udp.c (revision b0148a98ec5151fec82064d95f11eb9efbc628ea)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		The User Datagram Protocol (UDP).
7  *
8  * Version:	$Id: udp.c,v 1.102 2002/02/01 22:01:04 davem Exp $
9  *
10  * Authors:	Ross Biro
11  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12  *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
13  *		Alan Cox, <Alan.Cox@linux.org>
14  *		Hirokazu Takahashi, <taka@valinux.co.jp>
15  *
16  * Fixes:
17  *		Alan Cox	:	verify_area() calls
18  *		Alan Cox	: 	stopped close while in use off icmp
19  *					messages. Not a fix but a botch that
20  *					for udp at least is 'valid'.
21  *		Alan Cox	:	Fixed icmp handling properly
22  *		Alan Cox	: 	Correct error for oversized datagrams
23  *		Alan Cox	:	Tidied select() semantics.
24  *		Alan Cox	:	udp_err() fixed properly, also now
25  *					select and read wake correctly on errors
26  *		Alan Cox	:	udp_send verify_area moved to avoid mem leak
27  *		Alan Cox	:	UDP can count its memory
28  *		Alan Cox	:	send to an unknown connection causes
29  *					an ECONNREFUSED off the icmp, but
30  *					does NOT close.
31  *		Alan Cox	:	Switched to new sk_buff handlers. No more backlog!
32  *		Alan Cox	:	Using generic datagram code. Even smaller and the PEEK
33  *					bug no longer crashes it.
34  *		Fred Van Kempen	: 	Net2e support for sk->broadcast.
35  *		Alan Cox	:	Uses skb_free_datagram
36  *		Alan Cox	:	Added get/set sockopt support.
37  *		Alan Cox	:	Broadcasting without option set returns EACCES.
38  *		Alan Cox	:	No wakeup calls. Instead we now use the callbacks.
39  *		Alan Cox	:	Use ip_tos and ip_ttl
40  *		Alan Cox	:	SNMP Mibs
41  *		Alan Cox	:	MSG_DONTROUTE, and 0.0.0.0 support.
42  *		Matt Dillon	:	UDP length checks.
43  *		Alan Cox	:	Smarter af_inet used properly.
44  *		Alan Cox	:	Use new kernel side addressing.
45  *		Alan Cox	:	Incorrect return on truncated datagram receive.
46  *	Arnt Gulbrandsen 	:	New udp_send and stuff
47  *		Alan Cox	:	Cache last socket
48  *		Alan Cox	:	Route cache
49  *		Jon Peatfield	:	Minor efficiency fix to sendto().
50  *		Mike Shaver	:	RFC1122 checks.
51  *		Alan Cox	:	Nonblocking error fix.
52  *	Willy Konynenberg	:	Transparent proxying support.
53  *		Mike McLagan	:	Routing by source
54  *		David S. Miller	:	New socket lookup architecture.
55  *					Last socket cache retained as it
56  *					does have a high hit rate.
57  *		Olaf Kirch	:	Don't linearise iovec on sendmsg.
58  *		Andi Kleen	:	Some cleanups, cache destination entry
59  *					for connect.
60  *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
61  *		Melvin Smith	:	Check msg_name not msg_namelen in sendto(),
62  *					return ENOTCONN for unconnected sockets (POSIX)
63  *		Janos Farkas	:	don't deliver multi/broadcasts to a different
64  *					bound-to-device socket
65  *	Hirokazu Takahashi	:	HW checksumming for outgoing UDP
66  *					datagrams.
67  *	Hirokazu Takahashi	:	sendfile() on UDP works now.
68  *		Arnaldo C. Melo :	convert /proc/net/udp to seq_file
69  *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
70  *	Alexey Kuznetsov:		allow both IPv4 and IPv6 sockets to bind
71  *					a single port at the same time.
72  *	Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
73  *
74  *
75  *		This program is free software; you can redistribute it and/or
76  *		modify it under the terms of the GNU General Public License
77  *		as published by the Free Software Foundation; either version
78  *		2 of the License, or (at your option) any later version.
79  */
80 
81 #include <asm/system.h>
82 #include <asm/uaccess.h>
83 #include <asm/ioctls.h>
84 #include <linux/types.h>
85 #include <linux/fcntl.h>
86 #include <linux/module.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/igmp.h>
90 #include <linux/in.h>
91 #include <linux/errno.h>
92 #include <linux/timer.h>
93 #include <linux/mm.h>
94 #include <linux/inet.h>
95 #include <linux/netdevice.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/icmp.h>
101 #include <net/route.h>
102 #include <net/checksum.h>
103 #include <net/xfrm.h>
104 #include "udp_impl.h"
105 
106 /*
107  *	Snmp MIB for the UDP layer
108  */
109 
110 DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly;
111 
112 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
113 DEFINE_RWLOCK(udp_hash_lock);
114 
115 static int udp_port_rover;
116 
117 static inline int __udp_lib_lport_inuse(__u16 num, struct hlist_head udptable[])
118 {
119 	struct sock *sk;
120 	struct hlist_node *node;
121 
122 	sk_for_each(sk, node, &udptable[num & (UDP_HTABLE_SIZE - 1)])
123 		if (inet_sk(sk)->num == num)
124 			return 1;
125 	return 0;
126 }
127 
128 /**
129  *  __udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
130  *
131  *  @sk:          socket struct in question
132  *  @snum:        port number to look up
133  *  @udptable:    hash list table, must be of UDP_HTABLE_SIZE
134  *  @port_rover:  pointer to record of last unallocated port
135  *  @saddr_comp:  AF-dependent comparison of bound local IP addresses
136  */
137 int __udp_lib_get_port(struct sock *sk, unsigned short snum,
138 		       struct hlist_head udptable[], int *port_rover,
139 		       int (*saddr_comp)(const struct sock *sk1,
140 					 const struct sock *sk2 )    )
141 {
142 	struct hlist_node *node;
143 	struct hlist_head *head;
144 	struct sock *sk2;
145 	int    error = 1;
146 
147 	write_lock_bh(&udp_hash_lock);
148 	if (snum == 0) {
149 		int best_size_so_far, best, result, i;
150 
151 		if (*port_rover > sysctl_local_port_range[1] ||
152 		    *port_rover < sysctl_local_port_range[0])
153 			*port_rover = sysctl_local_port_range[0];
154 		best_size_so_far = 32767;
155 		best = result = *port_rover;
156 		for (i = 0; i < UDP_HTABLE_SIZE; i++, result++) {
157 			int size;
158 
159 			head = &udptable[result & (UDP_HTABLE_SIZE - 1)];
160 			if (hlist_empty(head)) {
161 				if (result > sysctl_local_port_range[1])
162 					result = sysctl_local_port_range[0] +
163 						((result - sysctl_local_port_range[0]) &
164 						 (UDP_HTABLE_SIZE - 1));
165 				goto gotit;
166 			}
167 			size = 0;
168 			sk_for_each(sk2, node, head) {
169 				if (++size >= best_size_so_far)
170 					goto next;
171 			}
172 			best_size_so_far = size;
173 			best = result;
174 		next:
175 			;
176 		}
177 		result = best;
178 		for(i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++, result += UDP_HTABLE_SIZE) {
179 			if (result > sysctl_local_port_range[1])
180 				result = sysctl_local_port_range[0]
181 					+ ((result - sysctl_local_port_range[0]) &
182 					   (UDP_HTABLE_SIZE - 1));
183 			if (! __udp_lib_lport_inuse(result, udptable))
184 				break;
185 		}
186 		if (i >= (1 << 16) / UDP_HTABLE_SIZE)
187 			goto fail;
188 gotit:
189 		*port_rover = snum = result;
190 	} else {
191 		head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
192 
193 		sk_for_each(sk2, node, head)
194 			if (inet_sk(sk2)->num == snum                        &&
195 			    sk2 != sk                                        &&
196 			    (!sk2->sk_reuse        || !sk->sk_reuse)         &&
197 			    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
198 			     || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
199 			    (*saddr_comp)(sk, sk2)                             )
200 				goto fail;
201 	}
202 	inet_sk(sk)->num = snum;
203 	if (sk_unhashed(sk)) {
204 		head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
205 		sk_add_node(sk, head);
206 		sock_prot_inc_use(sk->sk_prot);
207 	}
208 	error = 0;
209 fail:
210 	write_unlock_bh(&udp_hash_lock);
211 	return error;
212 }
213 
214 __inline__ int udp_get_port(struct sock *sk, unsigned short snum,
215 			int (*scmp)(const struct sock *, const struct sock *))
216 {
217 	return  __udp_lib_get_port(sk, snum, udp_hash, &udp_port_rover, scmp);
218 }
219 
220 inline int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
221 {
222 	struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
223 
224 	return 	( !ipv6_only_sock(sk2)  &&
225 		  (!inet1->rcv_saddr || !inet2->rcv_saddr ||
226 		   inet1->rcv_saddr == inet2->rcv_saddr      ));
227 }
228 
229 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum)
230 {
231 	return udp_get_port(sk, snum, ipv4_rcv_saddr_equal);
232 }
233 
234 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
235  * harder than this. -DaveM
236  */
237 static struct sock *__udp4_lib_lookup(__be32 saddr, __be16 sport,
238 				      __be32 daddr, __be16 dport,
239 				      int dif, struct hlist_head udptable[])
240 {
241 	struct sock *sk, *result = NULL;
242 	struct hlist_node *node;
243 	unsigned short hnum = ntohs(dport);
244 	int badness = -1;
245 
246 	read_lock(&udp_hash_lock);
247 	sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
248 		struct inet_sock *inet = inet_sk(sk);
249 
250 		if (inet->num == hnum && !ipv6_only_sock(sk)) {
251 			int score = (sk->sk_family == PF_INET ? 1 : 0);
252 			if (inet->rcv_saddr) {
253 				if (inet->rcv_saddr != daddr)
254 					continue;
255 				score+=2;
256 			}
257 			if (inet->daddr) {
258 				if (inet->daddr != saddr)
259 					continue;
260 				score+=2;
261 			}
262 			if (inet->dport) {
263 				if (inet->dport != sport)
264 					continue;
265 				score+=2;
266 			}
267 			if (sk->sk_bound_dev_if) {
268 				if (sk->sk_bound_dev_if != dif)
269 					continue;
270 				score+=2;
271 			}
272 			if(score == 9) {
273 				result = sk;
274 				break;
275 			} else if(score > badness) {
276 				result = sk;
277 				badness = score;
278 			}
279 		}
280 	}
281 	if (result)
282 		sock_hold(result);
283 	read_unlock(&udp_hash_lock);
284 	return result;
285 }
286 
287 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
288 					     __be16 loc_port, __be32 loc_addr,
289 					     __be16 rmt_port, __be32 rmt_addr,
290 					     int dif)
291 {
292 	struct hlist_node *node;
293 	struct sock *s = sk;
294 	unsigned short hnum = ntohs(loc_port);
295 
296 	sk_for_each_from(s, node) {
297 		struct inet_sock *inet = inet_sk(s);
298 
299 		if (inet->num != hnum					||
300 		    (inet->daddr && inet->daddr != rmt_addr)		||
301 		    (inet->dport != rmt_port && inet->dport)		||
302 		    (inet->rcv_saddr && inet->rcv_saddr != loc_addr)	||
303 		    ipv6_only_sock(s)					||
304 		    (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
305 			continue;
306 		if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
307 			continue;
308 		goto found;
309   	}
310 	s = NULL;
311 found:
312   	return s;
313 }
314 
315 /*
316  * This routine is called by the ICMP module when it gets some
317  * sort of error condition.  If err < 0 then the socket should
318  * be closed and the error returned to the user.  If err > 0
319  * it's just the icmp type << 8 | icmp code.
320  * Header points to the ip header of the error packet. We move
321  * on past this. Then (as it used to claim before adjustment)
322  * header points to the first 8 bytes of the udp header.  We need
323  * to find the appropriate port.
324  */
325 
326 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
327 {
328 	struct inet_sock *inet;
329 	struct iphdr *iph = (struct iphdr*)skb->data;
330 	struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
331 	int type = skb->h.icmph->type;
332 	int code = skb->h.icmph->code;
333 	struct sock *sk;
334 	int harderr;
335 	int err;
336 
337 	sk = __udp4_lib_lookup(iph->daddr, uh->dest, iph->saddr, uh->source,
338 			       skb->dev->ifindex, udptable		    );
339 	if (sk == NULL) {
340 		ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
341     	  	return;	/* No socket for error */
342 	}
343 
344 	err = 0;
345 	harderr = 0;
346 	inet = inet_sk(sk);
347 
348 	switch (type) {
349 	default:
350 	case ICMP_TIME_EXCEEDED:
351 		err = EHOSTUNREACH;
352 		break;
353 	case ICMP_SOURCE_QUENCH:
354 		goto out;
355 	case ICMP_PARAMETERPROB:
356 		err = EPROTO;
357 		harderr = 1;
358 		break;
359 	case ICMP_DEST_UNREACH:
360 		if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
361 			if (inet->pmtudisc != IP_PMTUDISC_DONT) {
362 				err = EMSGSIZE;
363 				harderr = 1;
364 				break;
365 			}
366 			goto out;
367 		}
368 		err = EHOSTUNREACH;
369 		if (code <= NR_ICMP_UNREACH) {
370 			harderr = icmp_err_convert[code].fatal;
371 			err = icmp_err_convert[code].errno;
372 		}
373 		break;
374 	}
375 
376 	/*
377 	 *      RFC1122: OK.  Passes ICMP errors back to application, as per
378 	 *	4.1.3.3.
379 	 */
380 	if (!inet->recverr) {
381 		if (!harderr || sk->sk_state != TCP_ESTABLISHED)
382 			goto out;
383 	} else {
384 		ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
385 	}
386 	sk->sk_err = err;
387 	sk->sk_error_report(sk);
388 out:
389 	sock_put(sk);
390 }
391 
392 __inline__ void udp_err(struct sk_buff *skb, u32 info)
393 {
394 	return __udp4_lib_err(skb, info, udp_hash);
395 }
396 
397 /*
398  * Throw away all pending data and cancel the corking. Socket is locked.
399  */
400 static void udp_flush_pending_frames(struct sock *sk)
401 {
402 	struct udp_sock *up = udp_sk(sk);
403 
404 	if (up->pending) {
405 		up->len = 0;
406 		up->pending = 0;
407 		ip_flush_pending_frames(sk);
408 	}
409 }
410 
411 /**
412  * 	udp4_hwcsum_outgoing  -  handle outgoing HW checksumming
413  * 	@sk: 	socket we are sending on
414  * 	@skb: 	sk_buff containing the filled-in UDP header
415  * 	        (checksum field must be zeroed out)
416  */
417 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
418 				 __be32 src, __be32 dst, int len      )
419 {
420 	unsigned int offset;
421 	struct udphdr *uh = skb->h.uh;
422 	__wsum csum = 0;
423 
424 	if (skb_queue_len(&sk->sk_write_queue) == 1) {
425 		/*
426 		 * Only one fragment on the socket.
427 		 */
428 		skb->csum_offset = offsetof(struct udphdr, check);
429 		uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
430 	} else {
431 		/*
432 		 * HW-checksum won't work as there are two or more
433 		 * fragments on the socket so that all csums of sk_buffs
434 		 * should be together
435 		 */
436 		offset = skb->h.raw - skb->data;
437 		skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
438 
439 		skb->ip_summed = CHECKSUM_NONE;
440 
441 		skb_queue_walk(&sk->sk_write_queue, skb) {
442 			csum = csum_add(csum, skb->csum);
443 		}
444 
445 		uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
446 		if (uh->check == 0)
447 			uh->check = CSUM_MANGLED_0;
448 	}
449 }
450 
451 /*
452  * Push out all pending data as one UDP datagram. Socket is locked.
453  */
454 static int udp_push_pending_frames(struct sock *sk)
455 {
456 	struct udp_sock  *up = udp_sk(sk);
457 	struct inet_sock *inet = inet_sk(sk);
458 	struct flowi *fl = &inet->cork.fl;
459 	struct sk_buff *skb;
460 	struct udphdr *uh;
461 	int err = 0;
462 	__wsum csum = 0;
463 
464 	/* Grab the skbuff where UDP header space exists. */
465 	if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
466 		goto out;
467 
468 	/*
469 	 * Create a UDP header
470 	 */
471 	uh = skb->h.uh;
472 	uh->source = fl->fl_ip_sport;
473 	uh->dest = fl->fl_ip_dport;
474 	uh->len = htons(up->len);
475 	uh->check = 0;
476 
477 	if (up->pcflag)  				 /*     UDP-Lite      */
478 		csum  = udplite_csum_outgoing(sk, skb);
479 
480 	else if (sk->sk_no_check == UDP_CSUM_NOXMIT) {   /* UDP csum disabled */
481 
482 		skb->ip_summed = CHECKSUM_NONE;
483 		goto send;
484 
485 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
486 
487 		udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
488 		goto send;
489 
490 	} else						 /*   `normal' UDP    */
491 		csum = udp_csum_outgoing(sk, skb);
492 
493 	/* add protocol-dependent pseudo-header */
494 	uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
495 				      sk->sk_protocol, csum             );
496 	if (uh->check == 0)
497 		uh->check = CSUM_MANGLED_0;
498 
499 send:
500 	err = ip_push_pending_frames(sk);
501 out:
502 	up->len = 0;
503 	up->pending = 0;
504 	return err;
505 }
506 
507 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
508 		size_t len)
509 {
510 	struct inet_sock *inet = inet_sk(sk);
511 	struct udp_sock *up = udp_sk(sk);
512 	int ulen = len;
513 	struct ipcm_cookie ipc;
514 	struct rtable *rt = NULL;
515 	int free = 0;
516 	int connected = 0;
517 	__be32 daddr, faddr, saddr;
518 	__be16 dport;
519 	u8  tos;
520 	int err, is_udplite = up->pcflag;
521 	int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
522 	int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
523 
524 	if (len > 0xFFFF)
525 		return -EMSGSIZE;
526 
527 	/*
528 	 *	Check the flags.
529 	 */
530 
531 	if (msg->msg_flags&MSG_OOB)	/* Mirror BSD error message compatibility */
532 		return -EOPNOTSUPP;
533 
534 	ipc.opt = NULL;
535 
536 	if (up->pending) {
537 		/*
538 		 * There are pending frames.
539 	 	 * The socket lock must be held while it's corked.
540 		 */
541 		lock_sock(sk);
542 		if (likely(up->pending)) {
543 			if (unlikely(up->pending != AF_INET)) {
544 				release_sock(sk);
545 				return -EINVAL;
546 			}
547  			goto do_append_data;
548 		}
549 		release_sock(sk);
550 	}
551 	ulen += sizeof(struct udphdr);
552 
553 	/*
554 	 *	Get and verify the address.
555 	 */
556 	if (msg->msg_name) {
557 		struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
558 		if (msg->msg_namelen < sizeof(*usin))
559 			return -EINVAL;
560 		if (usin->sin_family != AF_INET) {
561 			if (usin->sin_family != AF_UNSPEC)
562 				return -EAFNOSUPPORT;
563 		}
564 
565 		daddr = usin->sin_addr.s_addr;
566 		dport = usin->sin_port;
567 		if (dport == 0)
568 			return -EINVAL;
569 	} else {
570 		if (sk->sk_state != TCP_ESTABLISHED)
571 			return -EDESTADDRREQ;
572 		daddr = inet->daddr;
573 		dport = inet->dport;
574 		/* Open fast path for connected socket.
575 		   Route will not be used, if at least one option is set.
576 		 */
577 		connected = 1;
578   	}
579 	ipc.addr = inet->saddr;
580 
581 	ipc.oif = sk->sk_bound_dev_if;
582 	if (msg->msg_controllen) {
583 		err = ip_cmsg_send(msg, &ipc);
584 		if (err)
585 			return err;
586 		if (ipc.opt)
587 			free = 1;
588 		connected = 0;
589 	}
590 	if (!ipc.opt)
591 		ipc.opt = inet->opt;
592 
593 	saddr = ipc.addr;
594 	ipc.addr = faddr = daddr;
595 
596 	if (ipc.opt && ipc.opt->srr) {
597 		if (!daddr)
598 			return -EINVAL;
599 		faddr = ipc.opt->faddr;
600 		connected = 0;
601 	}
602 	tos = RT_TOS(inet->tos);
603 	if (sock_flag(sk, SOCK_LOCALROUTE) ||
604 	    (msg->msg_flags & MSG_DONTROUTE) ||
605 	    (ipc.opt && ipc.opt->is_strictroute)) {
606 		tos |= RTO_ONLINK;
607 		connected = 0;
608 	}
609 
610 	if (MULTICAST(daddr)) {
611 		if (!ipc.oif)
612 			ipc.oif = inet->mc_index;
613 		if (!saddr)
614 			saddr = inet->mc_addr;
615 		connected = 0;
616 	}
617 
618 	if (connected)
619 		rt = (struct rtable*)sk_dst_check(sk, 0);
620 
621 	if (rt == NULL) {
622 		struct flowi fl = { .oif = ipc.oif,
623 				    .nl_u = { .ip4_u =
624 					      { .daddr = faddr,
625 						.saddr = saddr,
626 						.tos = tos } },
627 				    .proto = sk->sk_protocol,
628 				    .uli_u = { .ports =
629 					       { .sport = inet->sport,
630 						 .dport = dport } } };
631 		security_sk_classify_flow(sk, &fl);
632 		err = ip_route_output_flow(&rt, &fl, sk, !(msg->msg_flags&MSG_DONTWAIT));
633 		if (err)
634 			goto out;
635 
636 		err = -EACCES;
637 		if ((rt->rt_flags & RTCF_BROADCAST) &&
638 		    !sock_flag(sk, SOCK_BROADCAST))
639 			goto out;
640 		if (connected)
641 			sk_dst_set(sk, dst_clone(&rt->u.dst));
642 	}
643 
644 	if (msg->msg_flags&MSG_CONFIRM)
645 		goto do_confirm;
646 back_from_confirm:
647 
648 	saddr = rt->rt_src;
649 	if (!ipc.addr)
650 		daddr = ipc.addr = rt->rt_dst;
651 
652 	lock_sock(sk);
653 	if (unlikely(up->pending)) {
654 		/* The socket is already corked while preparing it. */
655 		/* ... which is an evident application bug. --ANK */
656 		release_sock(sk);
657 
658 		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
659 		err = -EINVAL;
660 		goto out;
661 	}
662 	/*
663 	 *	Now cork the socket to pend data.
664 	 */
665 	inet->cork.fl.fl4_dst = daddr;
666 	inet->cork.fl.fl_ip_dport = dport;
667 	inet->cork.fl.fl4_src = saddr;
668 	inet->cork.fl.fl_ip_sport = inet->sport;
669 	up->pending = AF_INET;
670 
671 do_append_data:
672 	up->len += ulen;
673 	getfrag  =  is_udplite ?  udplite_getfrag : ip_generic_getfrag;
674 	err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
675 			sizeof(struct udphdr), &ipc, rt,
676 			corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
677 	if (err)
678 		udp_flush_pending_frames(sk);
679 	else if (!corkreq)
680 		err = udp_push_pending_frames(sk);
681 	else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
682 		up->pending = 0;
683 	release_sock(sk);
684 
685 out:
686 	ip_rt_put(rt);
687 	if (free)
688 		kfree(ipc.opt);
689 	if (!err) {
690 		UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
691 		return len;
692 	}
693 	/*
694 	 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
695 	 * ENOBUFS might not be good (it's not tunable per se), but otherwise
696 	 * we don't have a good statistic (IpOutDiscards but it can be too many
697 	 * things).  We could add another new stat but at least for now that
698 	 * seems like overkill.
699 	 */
700 	if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
701 		UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
702 	}
703 	return err;
704 
705 do_confirm:
706 	dst_confirm(&rt->u.dst);
707 	if (!(msg->msg_flags&MSG_PROBE) || len)
708 		goto back_from_confirm;
709 	err = 0;
710 	goto out;
711 }
712 
713 int udp_sendpage(struct sock *sk, struct page *page, int offset,
714 		 size_t size, int flags)
715 {
716 	struct udp_sock *up = udp_sk(sk);
717 	int ret;
718 
719 	if (!up->pending) {
720 		struct msghdr msg = {	.msg_flags = flags|MSG_MORE };
721 
722 		/* Call udp_sendmsg to specify destination address which
723 		 * sendpage interface can't pass.
724 		 * This will succeed only when the socket is connected.
725 		 */
726 		ret = udp_sendmsg(NULL, sk, &msg, 0);
727 		if (ret < 0)
728 			return ret;
729 	}
730 
731 	lock_sock(sk);
732 
733 	if (unlikely(!up->pending)) {
734 		release_sock(sk);
735 
736 		LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
737 		return -EINVAL;
738 	}
739 
740 	ret = ip_append_page(sk, page, offset, size, flags);
741 	if (ret == -EOPNOTSUPP) {
742 		release_sock(sk);
743 		return sock_no_sendpage(sk->sk_socket, page, offset,
744 					size, flags);
745 	}
746 	if (ret < 0) {
747 		udp_flush_pending_frames(sk);
748 		goto out;
749 	}
750 
751 	up->len += size;
752 	if (!(up->corkflag || (flags&MSG_MORE)))
753 		ret = udp_push_pending_frames(sk);
754 	if (!ret)
755 		ret = size;
756 out:
757 	release_sock(sk);
758 	return ret;
759 }
760 
761 /*
762  *	IOCTL requests applicable to the UDP protocol
763  */
764 
765 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
766 {
767 	switch(cmd)
768 	{
769 		case SIOCOUTQ:
770 		{
771 			int amount = atomic_read(&sk->sk_wmem_alloc);
772 			return put_user(amount, (int __user *)arg);
773 		}
774 
775 		case SIOCINQ:
776 		{
777 			struct sk_buff *skb;
778 			unsigned long amount;
779 
780 			amount = 0;
781 			spin_lock_bh(&sk->sk_receive_queue.lock);
782 			skb = skb_peek(&sk->sk_receive_queue);
783 			if (skb != NULL) {
784 				/*
785 				 * We will only return the amount
786 				 * of this packet since that is all
787 				 * that will be read.
788 				 */
789 				amount = skb->len - sizeof(struct udphdr);
790 			}
791 			spin_unlock_bh(&sk->sk_receive_queue.lock);
792 			return put_user(amount, (int __user *)arg);
793 		}
794 
795 		default:
796 			return -ENOIOCTLCMD;
797 	}
798 	return(0);
799 }
800 
801 /*
802  * 	This should be easy, if there is something there we
803  * 	return it, otherwise we block.
804  */
805 
806 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
807 	        size_t len, int noblock, int flags, int *addr_len)
808 {
809 	struct inet_sock *inet = inet_sk(sk);
810   	struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
811   	struct sk_buff *skb;
812 	int copied, err, copy_only, is_udplite = IS_UDPLITE(sk);
813 
814 	/*
815 	 *	Check any passed addresses
816 	 */
817 	if (addr_len)
818 		*addr_len=sizeof(*sin);
819 
820 	if (flags & MSG_ERRQUEUE)
821 		return ip_recv_error(sk, msg, len);
822 
823 try_again:
824 	skb = skb_recv_datagram(sk, flags, noblock, &err);
825 	if (!skb)
826 		goto out;
827 
828   	copied = skb->len - sizeof(struct udphdr);
829 	if (copied > len) {
830 		copied = len;
831 		msg->msg_flags |= MSG_TRUNC;
832 	}
833 
834 	/*
835 	 * 	Decide whether to checksum and/or copy data.
836 	 *
837 	 * 	UDP:      checksum may have been computed in HW,
838 	 * 	          (re-)compute it if message is truncated.
839 	 * 	UDP-Lite: always needs to checksum, no HW support.
840 	 */
841 	copy_only = (skb->ip_summed==CHECKSUM_UNNECESSARY);
842 
843 	if (is_udplite  ||  (!copy_only  &&  msg->msg_flags&MSG_TRUNC)) {
844 		if (__udp_lib_checksum_complete(skb))
845 			goto csum_copy_err;
846 		copy_only = 1;
847 	}
848 
849 	if (copy_only)
850 		err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
851 					      msg->msg_iov, copied       );
852 	else {
853 		err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
854 
855 		if (err == -EINVAL)
856 			goto csum_copy_err;
857 	}
858 
859 	if (err)
860 		goto out_free;
861 
862 	sock_recv_timestamp(msg, sk, skb);
863 
864 	/* Copy the address. */
865 	if (sin)
866 	{
867 		sin->sin_family = AF_INET;
868 		sin->sin_port = skb->h.uh->source;
869 		sin->sin_addr.s_addr = skb->nh.iph->saddr;
870 		memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
871   	}
872 	if (inet->cmsg_flags)
873 		ip_cmsg_recv(msg, skb);
874 
875 	err = copied;
876 	if (flags & MSG_TRUNC)
877 		err = skb->len - sizeof(struct udphdr);
878 
879 out_free:
880   	skb_free_datagram(sk, skb);
881 out:
882   	return err;
883 
884 csum_copy_err:
885 	UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
886 
887 	skb_kill_datagram(sk, skb, flags);
888 
889 	if (noblock)
890 		return -EAGAIN;
891 	goto try_again;
892 }
893 
894 
895 int udp_disconnect(struct sock *sk, int flags)
896 {
897 	struct inet_sock *inet = inet_sk(sk);
898 	/*
899 	 *	1003.1g - break association.
900 	 */
901 
902 	sk->sk_state = TCP_CLOSE;
903 	inet->daddr = 0;
904 	inet->dport = 0;
905 	sk->sk_bound_dev_if = 0;
906 	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
907 		inet_reset_saddr(sk);
908 
909 	if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
910 		sk->sk_prot->unhash(sk);
911 		inet->sport = 0;
912 	}
913 	sk_dst_reset(sk);
914 	return 0;
915 }
916 
917 /* return:
918  * 	1  if the the UDP system should process it
919  *	0  if we should drop this packet
920  * 	-1 if it should get processed by xfrm4_rcv_encap
921  */
922 static int udp_encap_rcv(struct sock * sk, struct sk_buff *skb)
923 {
924 #ifndef CONFIG_XFRM
925 	return 1;
926 #else
927 	struct udp_sock *up = udp_sk(sk);
928   	struct udphdr *uh;
929 	struct iphdr *iph;
930 	int iphlen, len;
931 
932 	__u8 *udpdata;
933 	__be32 *udpdata32;
934 	__u16 encap_type = up->encap_type;
935 
936 	/* if we're overly short, let UDP handle it */
937 	len = skb->len - sizeof(struct udphdr);
938 	if (len <= 0)
939 		return 1;
940 
941 	/* if this is not encapsulated socket, then just return now */
942 	if (!encap_type)
943 		return 1;
944 
945 	/* If this is a paged skb, make sure we pull up
946 	 * whatever data we need to look at. */
947 	if (!pskb_may_pull(skb, sizeof(struct udphdr) + min(len, 8)))
948 		return 1;
949 
950 	/* Now we can get the pointers */
951 	uh = skb->h.uh;
952 	udpdata = (__u8 *)uh + sizeof(struct udphdr);
953 	udpdata32 = (__be32 *)udpdata;
954 
955 	switch (encap_type) {
956 	default:
957 	case UDP_ENCAP_ESPINUDP:
958 		/* Check if this is a keepalive packet.  If so, eat it. */
959 		if (len == 1 && udpdata[0] == 0xff) {
960 			return 0;
961 		} else if (len > sizeof(struct ip_esp_hdr) && udpdata32[0] != 0 ) {
962 			/* ESP Packet without Non-ESP header */
963 			len = sizeof(struct udphdr);
964 		} else
965 			/* Must be an IKE packet.. pass it through */
966 			return 1;
967 		break;
968 	case UDP_ENCAP_ESPINUDP_NON_IKE:
969 		/* Check if this is a keepalive packet.  If so, eat it. */
970 		if (len == 1 && udpdata[0] == 0xff) {
971 			return 0;
972 		} else if (len > 2 * sizeof(u32) + sizeof(struct ip_esp_hdr) &&
973 			   udpdata32[0] == 0 && udpdata32[1] == 0) {
974 
975 			/* ESP Packet with Non-IKE marker */
976 			len = sizeof(struct udphdr) + 2 * sizeof(u32);
977 		} else
978 			/* Must be an IKE packet.. pass it through */
979 			return 1;
980 		break;
981 	}
982 
983 	/* At this point we are sure that this is an ESPinUDP packet,
984 	 * so we need to remove 'len' bytes from the packet (the UDP
985 	 * header and optional ESP marker bytes) and then modify the
986 	 * protocol to ESP, and then call into the transform receiver.
987 	 */
988 	if (skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
989 		return 0;
990 
991 	/* Now we can update and verify the packet length... */
992 	iph = skb->nh.iph;
993 	iphlen = iph->ihl << 2;
994 	iph->tot_len = htons(ntohs(iph->tot_len) - len);
995 	if (skb->len < iphlen + len) {
996 		/* packet is too small!?! */
997 		return 0;
998 	}
999 
1000 	/* pull the data buffer up to the ESP header and set the
1001 	 * transport header to point to ESP.  Keep UDP on the stack
1002 	 * for later.
1003 	 */
1004 	skb->h.raw = skb_pull(skb, len);
1005 
1006 	/* modify the protocol (it's ESP!) */
1007 	iph->protocol = IPPROTO_ESP;
1008 
1009 	/* and let the caller know to send this into the ESP processor... */
1010 	return -1;
1011 #endif
1012 }
1013 
1014 /* returns:
1015  *  -1: error
1016  *   0: success
1017  *  >0: "udp encap" protocol resubmission
1018  *
1019  * Note that in the success and error cases, the skb is assumed to
1020  * have either been requeued or freed.
1021  */
1022 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
1023 {
1024 	struct udp_sock *up = udp_sk(sk);
1025 	int rc;
1026 
1027 	/*
1028 	 *	Charge it to the socket, dropping if the queue is full.
1029 	 */
1030 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1031 		goto drop;
1032 	nf_reset(skb);
1033 
1034 	if (up->encap_type) {
1035 		/*
1036 		 * This is an encapsulation socket, so let's see if this is
1037 		 * an encapsulated packet.
1038 		 * If it's a keepalive packet, then just eat it.
1039 		 * If it's an encapsulateed packet, then pass it to the
1040 		 * IPsec xfrm input and return the response
1041 		 * appropriately.  Otherwise, just fall through and
1042 		 * pass this up the UDP socket.
1043 		 */
1044 		int ret;
1045 
1046 		ret = udp_encap_rcv(sk, skb);
1047 		if (ret == 0) {
1048 			/* Eat the packet .. */
1049 			kfree_skb(skb);
1050 			return 0;
1051 		}
1052 		if (ret < 0) {
1053 			/* process the ESP packet */
1054 			ret = xfrm4_rcv_encap(skb, up->encap_type);
1055 			UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
1056 			return -ret;
1057 		}
1058 		/* FALLTHROUGH -- it's a UDP Packet */
1059 	}
1060 
1061 	/*
1062 	 * 	UDP-Lite specific tests, ignored on UDP sockets
1063 	 */
1064 	if ((up->pcflag & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
1065 
1066 		/*
1067 		 * MIB statistics other than incrementing the error count are
1068 		 * disabled for the following two types of errors: these depend
1069 		 * on the application settings, not on the functioning of the
1070 		 * protocol stack as such.
1071 		 *
1072 		 * RFC 3828 here recommends (sec 3.3): "There should also be a
1073 		 * way ... to ... at least let the receiving application block
1074 		 * delivery of packets with coverage values less than a value
1075 		 * provided by the application."
1076 		 */
1077 		if (up->pcrlen == 0) {          /* full coverage was set  */
1078 			LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1079 				"%d while full coverage %d requested\n",
1080 				UDP_SKB_CB(skb)->cscov, skb->len);
1081 			goto drop;
1082 		}
1083 		/* The next case involves violating the min. coverage requested
1084 		 * by the receiver. This is subtle: if receiver wants x and x is
1085 		 * greater than the buffersize/MTU then receiver will complain
1086 		 * that it wants x while sender emits packets of smaller size y.
1087 		 * Therefore the above ...()->partial_cov statement is essential.
1088 		 */
1089 		if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
1090 			LIMIT_NETDEBUG(KERN_WARNING
1091 				"UDPLITE: coverage %d too small, need min %d\n",
1092 				UDP_SKB_CB(skb)->cscov, up->pcrlen);
1093 			goto drop;
1094 		}
1095 	}
1096 
1097 	if (sk->sk_filter && skb->ip_summed != CHECKSUM_UNNECESSARY) {
1098 		if (__udp_lib_checksum_complete(skb))
1099 			goto drop;
1100 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1101 	}
1102 
1103 	if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
1104 		/* Note that an ENOMEM error is charged twice */
1105 		if (rc == -ENOMEM)
1106 			UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, up->pcflag);
1107 		goto drop;
1108 	}
1109 
1110 	UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
1111 	return 0;
1112 
1113 drop:
1114 	UDP_INC_STATS_BH(UDP_MIB_INERRORS, up->pcflag);
1115 	kfree_skb(skb);
1116 	return -1;
1117 }
1118 
1119 /*
1120  *	Multicasts and broadcasts go to each listener.
1121  *
1122  *	Note: called only from the BH handler context,
1123  *	so we don't need to lock the hashes.
1124  */
1125 static int __udp4_lib_mcast_deliver(struct sk_buff *skb,
1126 				    struct udphdr  *uh,
1127 				    __be32 saddr, __be32 daddr,
1128 				    struct hlist_head udptable[])
1129 {
1130 	struct sock *sk;
1131 	int dif;
1132 
1133 	read_lock(&udp_hash_lock);
1134 	sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
1135 	dif = skb->dev->ifindex;
1136 	sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1137 	if (sk) {
1138 		struct sock *sknext = NULL;
1139 
1140 		do {
1141 			struct sk_buff *skb1 = skb;
1142 
1143 			sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1144 						   uh->source, saddr, dif);
1145 			if(sknext)
1146 				skb1 = skb_clone(skb, GFP_ATOMIC);
1147 
1148 			if(skb1) {
1149 				int ret = udp_queue_rcv_skb(sk, skb1);
1150 				if (ret > 0)
1151 					/* we should probably re-process instead
1152 					 * of dropping packets here. */
1153 					kfree_skb(skb1);
1154 			}
1155 			sk = sknext;
1156 		} while(sknext);
1157 	} else
1158 		kfree_skb(skb);
1159 	read_unlock(&udp_hash_lock);
1160 	return 0;
1161 }
1162 
1163 /* Initialize UDP checksum. If exited with zero value (success),
1164  * CHECKSUM_UNNECESSARY means, that no more checks are required.
1165  * Otherwise, csum completion requires chacksumming packet body,
1166  * including udp header and folding it to skb->csum.
1167  */
1168 static inline void udp4_csum_init(struct sk_buff *skb, struct udphdr *uh)
1169 {
1170 	if (uh->check == 0) {
1171 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1172 	} else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1173 	       if (!csum_tcpudp_magic(skb->nh.iph->saddr, skb->nh.iph->daddr,
1174 				      skb->len, IPPROTO_UDP, skb->csum       ))
1175 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1176 	}
1177 	if (skb->ip_summed != CHECKSUM_UNNECESSARY)
1178 		skb->csum = csum_tcpudp_nofold(skb->nh.iph->saddr,
1179 					       skb->nh.iph->daddr,
1180 					       skb->len, IPPROTO_UDP, 0);
1181 	/* Probably, we should checksum udp header (it should be in cache
1182 	 * in any case) and data in tiny packets (< rx copybreak).
1183 	 */
1184 
1185 	/* UDP = UDP-Lite with a non-partial checksum coverage */
1186 	UDP_SKB_CB(skb)->partial_cov = 0;
1187 }
1188 
1189 /*
1190  *	All we need to do is get the socket, and then do a checksum.
1191  */
1192 
1193 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1194 		   int is_udplite)
1195 {
1196   	struct sock *sk;
1197   	struct udphdr *uh = skb->h.uh;
1198 	unsigned short ulen;
1199 	struct rtable *rt = (struct rtable*)skb->dst;
1200 	__be32 saddr = skb->nh.iph->saddr;
1201 	__be32 daddr = skb->nh.iph->daddr;
1202 
1203 	/*
1204 	 *  Validate the packet.
1205 	 */
1206 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1207 		goto drop;		/* No space for header. */
1208 
1209 	ulen = ntohs(uh->len);
1210 	if (ulen > skb->len)
1211 		goto short_packet;
1212 
1213 	if(! is_udplite ) {		/* UDP validates ulen. */
1214 
1215 		if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1216 			goto short_packet;
1217 
1218 		udp4_csum_init(skb, uh);
1219 
1220 	} else 	{			/* UDP-Lite validates cscov. */
1221 		if (udplite4_csum_init(skb, uh))
1222 			goto csum_error;
1223 	}
1224 
1225 	if(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1226 		return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable);
1227 
1228 	sk = __udp4_lib_lookup(saddr, uh->source, daddr, uh->dest,
1229 			       skb->dev->ifindex, udptable        );
1230 
1231 	if (sk != NULL) {
1232 		int ret = udp_queue_rcv_skb(sk, skb);
1233 		sock_put(sk);
1234 
1235 		/* a return value > 0 means to resubmit the input, but
1236 		 * it wants the return to be -protocol, or 0
1237 		 */
1238 		if (ret > 0)
1239 			return -ret;
1240 		return 0;
1241 	}
1242 
1243 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1244 		goto drop;
1245 	nf_reset(skb);
1246 
1247 	/* No socket. Drop packet silently, if checksum is wrong */
1248 	if (udp_lib_checksum_complete(skb))
1249 		goto csum_error;
1250 
1251 	UDP_INC_STATS_BH(UDP_MIB_NOPORTS, is_udplite);
1252 	icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1253 
1254 	/*
1255 	 * Hmm.  We got an UDP packet to a port to which we
1256 	 * don't wanna listen.  Ignore it.
1257 	 */
1258 	kfree_skb(skb);
1259 	return(0);
1260 
1261 short_packet:
1262 	LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1263 		       is_udplite? "-Lite" : "",
1264 		       NIPQUAD(saddr),
1265 		       ntohs(uh->source),
1266 		       ulen,
1267 		       skb->len,
1268 		       NIPQUAD(daddr),
1269 		       ntohs(uh->dest));
1270 	goto drop;
1271 
1272 csum_error:
1273 	/*
1274 	 * RFC1122: OK.  Discards the bad packet silently (as far as
1275 	 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1276 	 */
1277 	LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1278 		       is_udplite? "-Lite" : "",
1279 		       NIPQUAD(saddr),
1280 		       ntohs(uh->source),
1281 		       NIPQUAD(daddr),
1282 		       ntohs(uh->dest),
1283 		       ulen);
1284 drop:
1285 	UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
1286 	kfree_skb(skb);
1287 	return(0);
1288 }
1289 
1290 __inline__ int udp_rcv(struct sk_buff *skb)
1291 {
1292 	return __udp4_lib_rcv(skb, udp_hash, 0);
1293 }
1294 
1295 int udp_destroy_sock(struct sock *sk)
1296 {
1297 	lock_sock(sk);
1298 	udp_flush_pending_frames(sk);
1299 	release_sock(sk);
1300 	return 0;
1301 }
1302 
1303 /*
1304  *	Socket option code for UDP
1305  */
1306 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1307 		       char __user *optval, int optlen,
1308 		       int (*push_pending_frames)(struct sock *))
1309 {
1310 	struct udp_sock *up = udp_sk(sk);
1311 	int val;
1312 	int err = 0;
1313 
1314 	if(optlen<sizeof(int))
1315 		return -EINVAL;
1316 
1317 	if (get_user(val, (int __user *)optval))
1318 		return -EFAULT;
1319 
1320 	switch(optname) {
1321 	case UDP_CORK:
1322 		if (val != 0) {
1323 			up->corkflag = 1;
1324 		} else {
1325 			up->corkflag = 0;
1326 			lock_sock(sk);
1327 			(*push_pending_frames)(sk);
1328 			release_sock(sk);
1329 		}
1330 		break;
1331 
1332 	case UDP_ENCAP:
1333 		switch (val) {
1334 		case 0:
1335 		case UDP_ENCAP_ESPINUDP:
1336 		case UDP_ENCAP_ESPINUDP_NON_IKE:
1337 			up->encap_type = val;
1338 			break;
1339 		default:
1340 			err = -ENOPROTOOPT;
1341 			break;
1342 		}
1343 		break;
1344 
1345 	/*
1346 	 * 	UDP-Lite's partial checksum coverage (RFC 3828).
1347 	 */
1348 	/* The sender sets actual checksum coverage length via this option.
1349 	 * The case coverage > packet length is handled by send module. */
1350 	case UDPLITE_SEND_CSCOV:
1351 		if (!up->pcflag)         /* Disable the option on UDP sockets */
1352 			return -ENOPROTOOPT;
1353 		if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1354 			val = 8;
1355 		up->pcslen = val;
1356 		up->pcflag |= UDPLITE_SEND_CC;
1357 		break;
1358 
1359         /* The receiver specifies a minimum checksum coverage value. To make
1360          * sense, this should be set to at least 8 (as done below). If zero is
1361 	 * used, this again means full checksum coverage.                     */
1362 	case UDPLITE_RECV_CSCOV:
1363 		if (!up->pcflag)         /* Disable the option on UDP sockets */
1364 			return -ENOPROTOOPT;
1365 		if (val != 0 && val < 8) /* Avoid silly minimal values.       */
1366 			val = 8;
1367 		up->pcrlen = val;
1368 		up->pcflag |= UDPLITE_RECV_CC;
1369 		break;
1370 
1371 	default:
1372 		err = -ENOPROTOOPT;
1373 		break;
1374 	};
1375 
1376 	return err;
1377 }
1378 
1379 int udp_setsockopt(struct sock *sk, int level, int optname,
1380 		   char __user *optval, int optlen)
1381 {
1382 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1383 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1384 					  udp_push_pending_frames);
1385 	return ip_setsockopt(sk, level, optname, optval, optlen);
1386 }
1387 
1388 #ifdef CONFIG_COMPAT
1389 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1390 			  char __user *optval, int optlen)
1391 {
1392 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1393 		return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1394 					  udp_push_pending_frames);
1395 	return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1396 }
1397 #endif
1398 
1399 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1400 		       char __user *optval, int __user *optlen)
1401 {
1402 	struct udp_sock *up = udp_sk(sk);
1403 	int val, len;
1404 
1405 	if(get_user(len,optlen))
1406 		return -EFAULT;
1407 
1408 	len = min_t(unsigned int, len, sizeof(int));
1409 
1410 	if(len < 0)
1411 		return -EINVAL;
1412 
1413 	switch(optname) {
1414 	case UDP_CORK:
1415 		val = up->corkflag;
1416 		break;
1417 
1418 	case UDP_ENCAP:
1419 		val = up->encap_type;
1420 		break;
1421 
1422 	/* The following two cannot be changed on UDP sockets, the return is
1423 	 * always 0 (which corresponds to the full checksum coverage of UDP). */
1424 	case UDPLITE_SEND_CSCOV:
1425 		val = up->pcslen;
1426 		break;
1427 
1428 	case UDPLITE_RECV_CSCOV:
1429 		val = up->pcrlen;
1430 		break;
1431 
1432 	default:
1433 		return -ENOPROTOOPT;
1434 	};
1435 
1436   	if(put_user(len, optlen))
1437   		return -EFAULT;
1438 	if(copy_to_user(optval, &val,len))
1439 		return -EFAULT;
1440   	return 0;
1441 }
1442 
1443 int udp_getsockopt(struct sock *sk, int level, int optname,
1444 		   char __user *optval, int __user *optlen)
1445 {
1446 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1447 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1448 	return ip_getsockopt(sk, level, optname, optval, optlen);
1449 }
1450 
1451 #ifdef CONFIG_COMPAT
1452 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1453 				 char __user *optval, int __user *optlen)
1454 {
1455 	if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1456 		return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1457 	return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1458 }
1459 #endif
1460 /**
1461  * 	udp_poll - wait for a UDP event.
1462  *	@file - file struct
1463  *	@sock - socket
1464  *	@wait - poll table
1465  *
1466  *	This is same as datagram poll, except for the special case of
1467  *	blocking sockets. If application is using a blocking fd
1468  *	and a packet with checksum error is in the queue;
1469  *	then it could get return from select indicating data available
1470  *	but then block when reading it. Add special case code
1471  *	to work around these arguably broken applications.
1472  */
1473 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1474 {
1475 	unsigned int mask = datagram_poll(file, sock, wait);
1476 	struct sock *sk = sock->sk;
1477 	int 	is_lite = IS_UDPLITE(sk);
1478 
1479 	/* Check for false positives due to checksum errors */
1480 	if ( (mask & POLLRDNORM) &&
1481 	     !(file->f_flags & O_NONBLOCK) &&
1482 	     !(sk->sk_shutdown & RCV_SHUTDOWN)){
1483 		struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1484 		struct sk_buff *skb;
1485 
1486 		spin_lock_bh(&rcvq->lock);
1487 		while ((skb = skb_peek(rcvq)) != NULL) {
1488 			if (udp_lib_checksum_complete(skb)) {
1489 				UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite);
1490 				__skb_unlink(skb, rcvq);
1491 				kfree_skb(skb);
1492 			} else {
1493 				skb->ip_summed = CHECKSUM_UNNECESSARY;
1494 				break;
1495 			}
1496 		}
1497 		spin_unlock_bh(&rcvq->lock);
1498 
1499 		/* nothing to see, move along */
1500 		if (skb == NULL)
1501 			mask &= ~(POLLIN | POLLRDNORM);
1502 	}
1503 
1504 	return mask;
1505 
1506 }
1507 
1508 struct proto udp_prot = {
1509  	.name		   = "UDP",
1510 	.owner		   = THIS_MODULE,
1511 	.close		   = udp_lib_close,
1512 	.connect	   = ip4_datagram_connect,
1513 	.disconnect	   = udp_disconnect,
1514 	.ioctl		   = udp_ioctl,
1515 	.destroy	   = udp_destroy_sock,
1516 	.setsockopt	   = udp_setsockopt,
1517 	.getsockopt	   = udp_getsockopt,
1518 	.sendmsg	   = udp_sendmsg,
1519 	.recvmsg	   = udp_recvmsg,
1520 	.sendpage	   = udp_sendpage,
1521 	.backlog_rcv	   = udp_queue_rcv_skb,
1522 	.hash		   = udp_lib_hash,
1523 	.unhash		   = udp_lib_unhash,
1524 	.get_port	   = udp_v4_get_port,
1525 	.obj_size	   = sizeof(struct udp_sock),
1526 #ifdef CONFIG_COMPAT
1527 	.compat_setsockopt = compat_udp_setsockopt,
1528 	.compat_getsockopt = compat_udp_getsockopt,
1529 #endif
1530 };
1531 
1532 /* ------------------------------------------------------------------------ */
1533 #ifdef CONFIG_PROC_FS
1534 
1535 static struct sock *udp_get_first(struct seq_file *seq)
1536 {
1537 	struct sock *sk;
1538 	struct udp_iter_state *state = seq->private;
1539 
1540 	for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1541 		struct hlist_node *node;
1542 		sk_for_each(sk, node, state->hashtable + state->bucket) {
1543 			if (sk->sk_family == state->family)
1544 				goto found;
1545 		}
1546 	}
1547 	sk = NULL;
1548 found:
1549 	return sk;
1550 }
1551 
1552 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1553 {
1554 	struct udp_iter_state *state = seq->private;
1555 
1556 	do {
1557 		sk = sk_next(sk);
1558 try_again:
1559 		;
1560 	} while (sk && sk->sk_family != state->family);
1561 
1562 	if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1563 		sk = sk_head(state->hashtable + state->bucket);
1564 		goto try_again;
1565 	}
1566 	return sk;
1567 }
1568 
1569 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1570 {
1571 	struct sock *sk = udp_get_first(seq);
1572 
1573 	if (sk)
1574 		while(pos && (sk = udp_get_next(seq, sk)) != NULL)
1575 			--pos;
1576 	return pos ? NULL : sk;
1577 }
1578 
1579 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1580 {
1581 	read_lock(&udp_hash_lock);
1582 	return *pos ? udp_get_idx(seq, *pos-1) : (void *)1;
1583 }
1584 
1585 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1586 {
1587 	struct sock *sk;
1588 
1589 	if (v == (void *)1)
1590 		sk = udp_get_idx(seq, 0);
1591 	else
1592 		sk = udp_get_next(seq, v);
1593 
1594 	++*pos;
1595 	return sk;
1596 }
1597 
1598 static void udp_seq_stop(struct seq_file *seq, void *v)
1599 {
1600 	read_unlock(&udp_hash_lock);
1601 }
1602 
1603 static int udp_seq_open(struct inode *inode, struct file *file)
1604 {
1605 	struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1606 	struct seq_file *seq;
1607 	int rc = -ENOMEM;
1608 	struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1609 
1610 	if (!s)
1611 		goto out;
1612 	s->family		= afinfo->family;
1613 	s->hashtable		= afinfo->hashtable;
1614 	s->seq_ops.start	= udp_seq_start;
1615 	s->seq_ops.next		= udp_seq_next;
1616 	s->seq_ops.show		= afinfo->seq_show;
1617 	s->seq_ops.stop		= udp_seq_stop;
1618 
1619 	rc = seq_open(file, &s->seq_ops);
1620 	if (rc)
1621 		goto out_kfree;
1622 
1623 	seq	     = file->private_data;
1624 	seq->private = s;
1625 out:
1626 	return rc;
1627 out_kfree:
1628 	kfree(s);
1629 	goto out;
1630 }
1631 
1632 /* ------------------------------------------------------------------------ */
1633 int udp_proc_register(struct udp_seq_afinfo *afinfo)
1634 {
1635 	struct proc_dir_entry *p;
1636 	int rc = 0;
1637 
1638 	if (!afinfo)
1639 		return -EINVAL;
1640 	afinfo->seq_fops->owner		= afinfo->owner;
1641 	afinfo->seq_fops->open		= udp_seq_open;
1642 	afinfo->seq_fops->read		= seq_read;
1643 	afinfo->seq_fops->llseek	= seq_lseek;
1644 	afinfo->seq_fops->release	= seq_release_private;
1645 
1646 	p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops);
1647 	if (p)
1648 		p->data = afinfo;
1649 	else
1650 		rc = -ENOMEM;
1651 	return rc;
1652 }
1653 
1654 void udp_proc_unregister(struct udp_seq_afinfo *afinfo)
1655 {
1656 	if (!afinfo)
1657 		return;
1658 	proc_net_remove(afinfo->name);
1659 	memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1660 }
1661 
1662 /* ------------------------------------------------------------------------ */
1663 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket)
1664 {
1665 	struct inet_sock *inet = inet_sk(sp);
1666 	__be32 dest = inet->daddr;
1667 	__be32 src  = inet->rcv_saddr;
1668 	__u16 destp	  = ntohs(inet->dport);
1669 	__u16 srcp	  = ntohs(inet->sport);
1670 
1671 	sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1672 		" %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1673 		bucket, src, srcp, dest, destp, sp->sk_state,
1674 		atomic_read(&sp->sk_wmem_alloc),
1675 		atomic_read(&sp->sk_rmem_alloc),
1676 		0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1677 		atomic_read(&sp->sk_refcnt), sp);
1678 }
1679 
1680 int udp4_seq_show(struct seq_file *seq, void *v)
1681 {
1682 	if (v == SEQ_START_TOKEN)
1683 		seq_printf(seq, "%-127s\n",
1684 			   "  sl  local_address rem_address   st tx_queue "
1685 			   "rx_queue tr tm->when retrnsmt   uid  timeout "
1686 			   "inode");
1687 	else {
1688 		char tmpbuf[129];
1689 		struct udp_iter_state *state = seq->private;
1690 
1691 		udp4_format_sock(v, tmpbuf, state->bucket);
1692 		seq_printf(seq, "%-127s\n", tmpbuf);
1693 	}
1694 	return 0;
1695 }
1696 
1697 /* ------------------------------------------------------------------------ */
1698 static struct file_operations udp4_seq_fops;
1699 static struct udp_seq_afinfo udp4_seq_afinfo = {
1700 	.owner		= THIS_MODULE,
1701 	.name		= "udp",
1702 	.family		= AF_INET,
1703 	.hashtable	= udp_hash,
1704 	.seq_show	= udp4_seq_show,
1705 	.seq_fops	= &udp4_seq_fops,
1706 };
1707 
1708 int __init udp4_proc_init(void)
1709 {
1710 	return udp_proc_register(&udp4_seq_afinfo);
1711 }
1712 
1713 void udp4_proc_exit(void)
1714 {
1715 	udp_proc_unregister(&udp4_seq_afinfo);
1716 }
1717 #endif /* CONFIG_PROC_FS */
1718 
1719 EXPORT_SYMBOL(udp_disconnect);
1720 EXPORT_SYMBOL(udp_hash);
1721 EXPORT_SYMBOL(udp_hash_lock);
1722 EXPORT_SYMBOL(udp_ioctl);
1723 EXPORT_SYMBOL(udp_get_port);
1724 EXPORT_SYMBOL(udp_prot);
1725 EXPORT_SYMBOL(udp_sendmsg);
1726 EXPORT_SYMBOL(udp_lib_getsockopt);
1727 EXPORT_SYMBOL(udp_lib_setsockopt);
1728 EXPORT_SYMBOL(udp_poll);
1729 
1730 #ifdef CONFIG_PROC_FS
1731 EXPORT_SYMBOL(udp_proc_register);
1732 EXPORT_SYMBOL(udp_proc_unregister);
1733 #endif
1734