xref: /freebsd/sys/netinet/udp_usrreq.c (revision c0020399a650364d0134f79f3fa319f84064372d)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.
4  * Copyright (c) 2008 Robert N. M. Watson
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 4. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)udp_usrreq.c	8.6 (Berkeley) 5/23/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_ipfw.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_mac.h"
41 
42 #include <sys/param.h>
43 #include <sys/domain.h>
44 #include <sys/eventhandler.h>
45 #include <sys/jail.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/protosw.h>
53 #include <sys/signalvar.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/sx.h>
57 #include <sys/sysctl.h>
58 #include <sys/syslog.h>
59 #include <sys/systm.h>
60 #include <sys/vimage.h>
61 
62 #include <vm/uma.h>
63 
64 #include <net/if.h>
65 #include <net/route.h>
66 
67 #include <netinet/in.h>
68 #include <netinet/in_pcb.h>
69 #include <netinet/in_systm.h>
70 #include <netinet/in_var.h>
71 #include <netinet/ip.h>
72 #ifdef INET6
73 #include <netinet/ip6.h>
74 #endif
75 #include <netinet/ip_icmp.h>
76 #include <netinet/icmp_var.h>
77 #include <netinet/ip_var.h>
78 #include <netinet/ip_options.h>
79 #ifdef INET6
80 #include <netinet6/ip6_var.h>
81 #endif
82 #include <netinet/udp.h>
83 #include <netinet/udp_var.h>
84 #include <netinet/vinet.h>
85 
86 #ifdef IPSEC
87 #include <netipsec/ipsec.h>
88 #endif
89 
90 #include <machine/in_cksum.h>
91 
92 #include <security/mac/mac_framework.h>
93 
94 /*
95  * UDP protocol implementation.
96  * Per RFC 768, August, 1980.
97  */
98 
99 #ifdef VIMAGE_GLOBALS
100 int	udp_blackhole;
101 #endif
102 
103 /*
104  * BSD 4.2 defaulted the udp checksum to be off.  Turning off udp checksums
105  * removes the only data integrity mechanism for packets and malformed
106  * packets that would otherwise be discarded due to bad checksums, and may
107  * cause problems (especially for NFS data blocks).
108  */
109 static int	udp_cksum = 1;
110 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, &udp_cksum,
111     0, "compute udp checksum");
112 
113 int	udp_log_in_vain = 0;
114 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW,
115     &udp_log_in_vain, 0, "Log all incoming UDP packets");
116 
117 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_udp, OID_AUTO, blackhole,
118     CTLFLAG_RW, udp_blackhole, 0,
119     "Do not send port unreachables for refused connects");
120 
121 u_long	udp_sendspace = 9216;		/* really max datagram size */
122 					/* 40 1K datagrams */
123 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW,
124     &udp_sendspace, 0, "Maximum outgoing UDP datagram size");
125 
126 u_long	udp_recvspace = 40 * (1024 +
127 #ifdef INET6
128 				      sizeof(struct sockaddr_in6)
129 #else
130 				      sizeof(struct sockaddr_in)
131 #endif
132 				      );
133 
134 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
135     &udp_recvspace, 0, "Maximum space for incoming UDP datagrams");
136 
137 #ifdef VIMAGE_GLOBALS
138 struct inpcbhead	udb;		/* from udp_var.h */
139 struct inpcbinfo	udbinfo;
140 struct udpstat		udpstat;	/* from udp_var.h */
141 #endif
142 
143 #ifndef UDBHASHSIZE
144 #define	UDBHASHSIZE	128
145 #endif
146 
147 SYSCTL_V_STRUCT(V_NET, vnet_inet, _net_inet_udp, UDPCTL_STATS, stats,
148     CTLFLAG_RW, udpstat, udpstat,
149     "UDP statistics (struct udpstat, netinet/udp_var.h)");
150 
151 static void	udp_detach(struct socket *so);
152 static int	udp_output(struct inpcb *, struct mbuf *, struct sockaddr *,
153 		    struct mbuf *, struct thread *);
154 
155 static void
156 udp_zone_change(void *tag)
157 {
158 
159 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
160 }
161 
162 static int
163 udp_inpcb_init(void *mem, int size, int flags)
164 {
165 	struct inpcb *inp;
166 
167 	inp = mem;
168 	INP_LOCK_INIT(inp, "inp", "udpinp");
169 	return (0);
170 }
171 
172 void
173 udp_init(void)
174 {
175 	INIT_VNET_INET(curvnet);
176 
177 	V_udp_blackhole = 0;
178 
179 	INP_INFO_LOCK_INIT(&V_udbinfo, "udp");
180 	LIST_INIT(&V_udb);
181 	V_udbinfo.ipi_listhead = &V_udb;
182 	V_udbinfo.ipi_hashbase = hashinit(UDBHASHSIZE, M_PCB,
183 	    &V_udbinfo.ipi_hashmask);
184 	V_udbinfo.ipi_porthashbase = hashinit(UDBHASHSIZE, M_PCB,
185 	    &V_udbinfo.ipi_porthashmask);
186 	V_udbinfo.ipi_zone = uma_zcreate("udpcb", sizeof(struct inpcb), NULL,
187 	    NULL, udp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
188 	uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets);
189 	EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL,
190 	    EVENTHANDLER_PRI_ANY);
191 }
192 
193 /*
194  * Subroutine of udp_input(), which appends the provided mbuf chain to the
195  * passed pcb/socket.  The caller must provide a sockaddr_in via udp_in that
196  * contains the source address.  If the socket ends up being an IPv6 socket,
197  * udp_append() will convert to a sockaddr_in6 before passing the address
198  * into the socket code.
199  */
200 static void
201 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
202     struct sockaddr_in *udp_in)
203 {
204 	struct sockaddr *append_sa;
205 	struct socket *so;
206 	struct mbuf *opts = 0;
207 #ifdef INET6
208 	struct sockaddr_in6 udp_in6;
209 #endif
210 
211 	INP_RLOCK_ASSERT(inp);
212 
213 #ifdef IPSEC
214 	/* Check AH/ESP integrity. */
215 	if (ipsec4_in_reject(n, inp)) {
216 		INIT_VNET_IPSEC(curvnet);
217 		m_freem(n);
218 		V_ipsec4stat.in_polvio++;
219 		return;
220 	}
221 #endif /* IPSEC */
222 #ifdef MAC
223 	if (mac_inpcb_check_deliver(inp, n) != 0) {
224 		m_freem(n);
225 		return;
226 	}
227 #endif
228 	if (inp->inp_flags & INP_CONTROLOPTS ||
229 	    inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) {
230 #ifdef INET6
231 		if (inp->inp_vflag & INP_IPV6)
232 			(void)ip6_savecontrol_v4(inp, n, &opts, NULL);
233 		else
234 #endif
235 			ip_savecontrol(inp, &opts, ip, n);
236 	}
237 #ifdef INET6
238 	if (inp->inp_vflag & INP_IPV6) {
239 		bzero(&udp_in6, sizeof(udp_in6));
240 		udp_in6.sin6_len = sizeof(udp_in6);
241 		udp_in6.sin6_family = AF_INET6;
242 		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
243 		append_sa = (struct sockaddr *)&udp_in6;
244 	} else
245 #endif
246 		append_sa = (struct sockaddr *)udp_in;
247 	m_adj(n, off);
248 
249 	so = inp->inp_socket;
250 	SOCKBUF_LOCK(&so->so_rcv);
251 	if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) {
252 		INIT_VNET_INET(so->so_vnet);
253 		SOCKBUF_UNLOCK(&so->so_rcv);
254 		m_freem(n);
255 		if (opts)
256 			m_freem(opts);
257 		UDPSTAT_INC(udps_fullsock);
258 	} else
259 		sorwakeup_locked(so);
260 }
261 
262 void
263 udp_input(struct mbuf *m, int off)
264 {
265 	INIT_VNET_INET(curvnet);
266 	int iphlen = off;
267 	struct ip *ip;
268 	struct udphdr *uh;
269 	struct ifnet *ifp;
270 	struct inpcb *inp;
271 	int len;
272 	struct ip save_ip;
273 	struct sockaddr_in udp_in;
274 #ifdef IPFIREWALL_FORWARD
275 	struct m_tag *fwd_tag;
276 #endif
277 
278 	ifp = m->m_pkthdr.rcvif;
279 	UDPSTAT_INC(udps_ipackets);
280 
281 	/*
282 	 * Strip IP options, if any; should skip this, make available to
283 	 * user, and use on returned packets, but we don't yet have a way to
284 	 * check the checksum with options still present.
285 	 */
286 	if (iphlen > sizeof (struct ip)) {
287 		ip_stripoptions(m, (struct mbuf *)0);
288 		iphlen = sizeof(struct ip);
289 	}
290 
291 	/*
292 	 * Get IP and UDP header together in first mbuf.
293 	 */
294 	ip = mtod(m, struct ip *);
295 	if (m->m_len < iphlen + sizeof(struct udphdr)) {
296 		if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == 0) {
297 			UDPSTAT_INC(udps_hdrops);
298 			return;
299 		}
300 		ip = mtod(m, struct ip *);
301 	}
302 	uh = (struct udphdr *)((caddr_t)ip + iphlen);
303 
304 	/*
305 	 * Destination port of 0 is illegal, based on RFC768.
306 	 */
307 	if (uh->uh_dport == 0)
308 		goto badunlocked;
309 
310 	/*
311 	 * Construct sockaddr format source address.  Stuff source address
312 	 * and datagram in user buffer.
313 	 */
314 	bzero(&udp_in, sizeof(udp_in));
315 	udp_in.sin_len = sizeof(udp_in);
316 	udp_in.sin_family = AF_INET;
317 	udp_in.sin_port = uh->uh_sport;
318 	udp_in.sin_addr = ip->ip_src;
319 
320 	/*
321 	 * Make mbuf data length reflect UDP length.  If not enough data to
322 	 * reflect UDP length, drop.
323 	 */
324 	len = ntohs((u_short)uh->uh_ulen);
325 	if (ip->ip_len != len) {
326 		if (len > ip->ip_len || len < sizeof(struct udphdr)) {
327 			UDPSTAT_INC(udps_badlen);
328 			goto badunlocked;
329 		}
330 		m_adj(m, len - ip->ip_len);
331 		/* ip->ip_len = len; */
332 	}
333 
334 	/*
335 	 * Save a copy of the IP header in case we want restore it for
336 	 * sending an ICMP error message in response.
337 	 */
338 	if (!V_udp_blackhole)
339 		save_ip = *ip;
340 	else
341 		memset(&save_ip, 0, sizeof(save_ip));
342 
343 	/*
344 	 * Checksum extended UDP header and data.
345 	 */
346 	if (uh->uh_sum) {
347 		u_short uh_sum;
348 
349 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
350 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
351 				uh_sum = m->m_pkthdr.csum_data;
352 			else
353 				uh_sum = in_pseudo(ip->ip_src.s_addr,
354 				    ip->ip_dst.s_addr, htonl((u_short)len +
355 				    m->m_pkthdr.csum_data + IPPROTO_UDP));
356 			uh_sum ^= 0xffff;
357 		} else {
358 			char b[9];
359 
360 			bcopy(((struct ipovly *)ip)->ih_x1, b, 9);
361 			bzero(((struct ipovly *)ip)->ih_x1, 9);
362 			((struct ipovly *)ip)->ih_len = uh->uh_ulen;
363 			uh_sum = in_cksum(m, len + sizeof (struct ip));
364 			bcopy(b, ((struct ipovly *)ip)->ih_x1, 9);
365 		}
366 		if (uh_sum) {
367 			UDPSTAT_INC(udps_badsum);
368 			m_freem(m);
369 			return;
370 		}
371 	} else
372 		UDPSTAT_INC(udps_nosum);
373 
374 #ifdef IPFIREWALL_FORWARD
375 	/*
376 	 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
377 	 */
378 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
379 	if (fwd_tag != NULL) {
380 		struct sockaddr_in *next_hop;
381 
382 		/*
383 		 * Do the hack.
384 		 */
385 		next_hop = (struct sockaddr_in *)(fwd_tag + 1);
386 		ip->ip_dst = next_hop->sin_addr;
387 		uh->uh_dport = ntohs(next_hop->sin_port);
388 
389 		/*
390 		 * Remove the tag from the packet.  We don't need it anymore.
391 		 */
392 		m_tag_delete(m, fwd_tag);
393 	}
394 #endif
395 
396 	INP_INFO_RLOCK(&V_udbinfo);
397 	if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
398 	    in_broadcast(ip->ip_dst, ifp)) {
399 		struct inpcb *last;
400 		struct ip_moptions *imo;
401 
402 		last = NULL;
403 		LIST_FOREACH(inp, &V_udb, inp_list) {
404 			if (inp->inp_lport != uh->uh_dport)
405 				continue;
406 #ifdef INET6
407 			if ((inp->inp_vflag & INP_IPV4) == 0)
408 				continue;
409 #endif
410 			if (inp->inp_laddr.s_addr != INADDR_ANY &&
411 			    inp->inp_laddr.s_addr != ip->ip_dst.s_addr)
412 				continue;
413 			if (inp->inp_faddr.s_addr != INADDR_ANY &&
414 			    inp->inp_faddr.s_addr != ip->ip_src.s_addr)
415 				continue;
416 			if (inp->inp_fport != 0 &&
417 			    inp->inp_fport != uh->uh_sport)
418 				continue;
419 
420 			INP_RLOCK(inp);
421 
422 			/*
423 			 * Handle socket delivery policy for any-source
424 			 * and source-specific multicast. [RFC3678]
425 			 */
426 			imo = inp->inp_moptions;
427 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
428 			    imo != NULL) {
429 				struct sockaddr_in	 group;
430 				int			 blocked;
431 
432 				bzero(&group, sizeof(struct sockaddr_in));
433 				group.sin_len = sizeof(struct sockaddr_in);
434 				group.sin_family = AF_INET;
435 				group.sin_addr = ip->ip_dst;
436 
437 				blocked = imo_multi_filter(imo, ifp,
438 					(struct sockaddr *)&group,
439 					(struct sockaddr *)&udp_in);
440 				if (blocked != MCAST_PASS) {
441 					if (blocked == MCAST_NOTGMEMBER)
442 						IPSTAT_INC(ips_notmember);
443 					if (blocked == MCAST_NOTSMEMBER ||
444 					    blocked == MCAST_MUTED)
445 						UDPSTAT_INC(udps_filtermcast);
446 					INP_RUNLOCK(inp);
447 					continue;
448 				}
449 			}
450 			if (last != NULL) {
451 				struct mbuf *n;
452 
453 				n = m_copy(m, 0, M_COPYALL);
454 				if (last->inp_ppcb == NULL) {
455 					if (n != NULL)
456 						udp_append(last,
457 						    ip, n,
458 						    iphlen +
459 						    sizeof(struct udphdr),
460 						    &udp_in);
461 					INP_RUNLOCK(last);
462 				} else {
463 					/*
464 					 * Engage the tunneling protocol we
465 					 * will have to leave the info_lock
466 					 * up, since we are hunting through
467 					 * multiple UDP's.
468 					 *
469 					 */
470 					udp_tun_func_t tunnel_func;
471 
472 					tunnel_func = (udp_tun_func_t)last->inp_ppcb;
473 					tunnel_func(n, iphlen, last);
474 					INP_RUNLOCK(last);
475 				}
476 			}
477 			last = inp;
478 			/*
479 			 * Don't look for additional matches if this one does
480 			 * not have either the SO_REUSEPORT or SO_REUSEADDR
481 			 * socket options set.  This heuristic avoids
482 			 * searching through all pcbs in the common case of a
483 			 * non-shared port.  It assumes that an application
484 			 * will never clear these options after setting them.
485 			 */
486 			if ((last->inp_socket->so_options &
487 			    (SO_REUSEPORT|SO_REUSEADDR)) == 0)
488 				break;
489 		}
490 
491 		if (last == NULL) {
492 			/*
493 			 * No matching pcb found; discard datagram.  (No need
494 			 * to send an ICMP Port Unreachable for a broadcast
495 			 * or multicast datgram.)
496 			 */
497 			UDPSTAT_INC(udps_noportbcast);
498 			goto badheadlocked;
499 		}
500 		if (last->inp_ppcb == NULL) {
501 			udp_append(last, ip, m, iphlen + sizeof(struct udphdr),
502 			    &udp_in);
503 			INP_RUNLOCK(last);
504 			INP_INFO_RUNLOCK(&V_udbinfo);
505 		} else {
506 			/*
507 			 * Engage the tunneling protocol.
508 			 */
509 			udp_tun_func_t tunnel_func;
510 
511 			tunnel_func = (udp_tun_func_t)last->inp_ppcb;
512 			tunnel_func(m, iphlen, last);
513 			INP_RUNLOCK(last);
514 			INP_INFO_RUNLOCK(&V_udbinfo);
515 		}
516 		return;
517 	}
518 
519 	/*
520 	 * Locate pcb for datagram.
521 	 */
522 	inp = in_pcblookup_hash(&V_udbinfo, ip->ip_src, uh->uh_sport,
523 	    ip->ip_dst, uh->uh_dport, 1, ifp);
524 	if (inp == NULL) {
525 		if (udp_log_in_vain) {
526 			char buf[4*sizeof "123"];
527 
528 			strcpy(buf, inet_ntoa(ip->ip_dst));
529 			log(LOG_INFO,
530 			    "Connection attempt to UDP %s:%d from %s:%d\n",
531 			    buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src),
532 			    ntohs(uh->uh_sport));
533 		}
534 		UDPSTAT_INC(udps_noport);
535 		if (m->m_flags & (M_BCAST | M_MCAST)) {
536 			UDPSTAT_INC(udps_noportbcast);
537 			goto badheadlocked;
538 		}
539 		if (V_udp_blackhole)
540 			goto badheadlocked;
541 		if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0)
542 			goto badheadlocked;
543 		*ip = save_ip;
544 		ip->ip_len += iphlen;
545 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0);
546 		INP_INFO_RUNLOCK(&V_udbinfo);
547 		return;
548 	}
549 
550 	/*
551 	 * Check the minimum TTL for socket.
552 	 */
553 	INP_RLOCK(inp);
554 	INP_INFO_RUNLOCK(&V_udbinfo);
555 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) {
556 		INP_RUNLOCK(inp);
557 		goto badunlocked;
558 	}
559 	if (inp->inp_ppcb != NULL) {
560 		/*
561 		 * Engage the tunneling protocol.
562 		 */
563 		udp_tun_func_t tunnel_func;
564 
565 		tunnel_func = (udp_tun_func_t)inp->inp_ppcb;
566 		tunnel_func(m, iphlen, inp);
567 		INP_RUNLOCK(inp);
568 		return;
569 	}
570 	udp_append(inp, ip, m, iphlen + sizeof(struct udphdr), &udp_in);
571 	INP_RUNLOCK(inp);
572 	return;
573 
574 badheadlocked:
575 	if (inp)
576 		INP_RUNLOCK(inp);
577 	INP_INFO_RUNLOCK(&V_udbinfo);
578 badunlocked:
579 	m_freem(m);
580 }
581 
582 /*
583  * Notify a udp user of an asynchronous error; just wake up so that they can
584  * collect error status.
585  */
586 struct inpcb *
587 udp_notify(struct inpcb *inp, int errno)
588 {
589 
590 	/*
591 	 * While udp_ctlinput() always calls udp_notify() with a read lock
592 	 * when invoking it directly, in_pcbnotifyall() currently uses write
593 	 * locks due to sharing code with TCP.  For now, accept either a read
594 	 * or a write lock, but a read lock is sufficient.
595 	 */
596 	INP_LOCK_ASSERT(inp);
597 
598 	inp->inp_socket->so_error = errno;
599 	sorwakeup(inp->inp_socket);
600 	sowwakeup(inp->inp_socket);
601 	return (inp);
602 }
603 
604 void
605 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
606 {
607 	INIT_VNET_INET(curvnet);
608 	struct ip *ip = vip;
609 	struct udphdr *uh;
610 	struct in_addr faddr;
611 	struct inpcb *inp;
612 
613 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
614 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
615 		return;
616 
617 	/*
618 	 * Redirects don't need to be handled up here.
619 	 */
620 	if (PRC_IS_REDIRECT(cmd))
621 		return;
622 
623 	/*
624 	 * Hostdead is ugly because it goes linearly through all PCBs.
625 	 *
626 	 * XXX: We never get this from ICMP, otherwise it makes an excellent
627 	 * DoS attack on machines with many connections.
628 	 */
629 	if (cmd == PRC_HOSTDEAD)
630 		ip = NULL;
631 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
632 		return;
633 	if (ip != NULL) {
634 		uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2));
635 		INP_INFO_RLOCK(&V_udbinfo);
636 		inp = in_pcblookup_hash(&V_udbinfo, faddr, uh->uh_dport,
637 		    ip->ip_src, uh->uh_sport, 0, NULL);
638 		if (inp != NULL) {
639 			INP_RLOCK(inp);
640 			if (inp->inp_socket != NULL) {
641 				udp_notify(inp, inetctlerrmap[cmd]);
642 			}
643 			INP_RUNLOCK(inp);
644 		}
645 		INP_INFO_RUNLOCK(&V_udbinfo);
646 	} else
647 		in_pcbnotifyall(&V_udbinfo, faddr, inetctlerrmap[cmd],
648 		    udp_notify);
649 }
650 
651 static int
652 udp_pcblist(SYSCTL_HANDLER_ARGS)
653 {
654 	INIT_VNET_INET(curvnet);
655 	int error, i, n;
656 	struct inpcb *inp, **inp_list;
657 	inp_gen_t gencnt;
658 	struct xinpgen xig;
659 
660 	/*
661 	 * The process of preparing the PCB list is too time-consuming and
662 	 * resource-intensive to repeat twice on every request.
663 	 */
664 	if (req->oldptr == 0) {
665 		n = V_udbinfo.ipi_count;
666 		req->oldidx = 2 * (sizeof xig)
667 			+ (n + n/8) * sizeof(struct xinpcb);
668 		return (0);
669 	}
670 
671 	if (req->newptr != 0)
672 		return (EPERM);
673 
674 	/*
675 	 * OK, now we're committed to doing something.
676 	 */
677 	INP_INFO_RLOCK(&V_udbinfo);
678 	gencnt = V_udbinfo.ipi_gencnt;
679 	n = V_udbinfo.ipi_count;
680 	INP_INFO_RUNLOCK(&V_udbinfo);
681 
682 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
683 		+ n * sizeof(struct xinpcb));
684 	if (error != 0)
685 		return (error);
686 
687 	xig.xig_len = sizeof xig;
688 	xig.xig_count = n;
689 	xig.xig_gen = gencnt;
690 	xig.xig_sogen = so_gencnt;
691 	error = SYSCTL_OUT(req, &xig, sizeof xig);
692 	if (error)
693 		return (error);
694 
695 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
696 	if (inp_list == 0)
697 		return (ENOMEM);
698 
699 	INP_INFO_RLOCK(&V_udbinfo);
700 	for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n;
701 	     inp = LIST_NEXT(inp, inp_list)) {
702 		INP_RLOCK(inp);
703 		if (inp->inp_gencnt <= gencnt &&
704 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0)
705 			inp_list[i++] = inp;
706 		INP_RUNLOCK(inp);
707 	}
708 	INP_INFO_RUNLOCK(&V_udbinfo);
709 	n = i;
710 
711 	error = 0;
712 	for (i = 0; i < n; i++) {
713 		inp = inp_list[i];
714 		INP_RLOCK(inp);
715 		if (inp->inp_gencnt <= gencnt) {
716 			struct xinpcb xi;
717 			bzero(&xi, sizeof(xi));
718 			xi.xi_len = sizeof xi;
719 			/* XXX should avoid extra copy */
720 			bcopy(inp, &xi.xi_inp, sizeof *inp);
721 			if (inp->inp_socket)
722 				sotoxsocket(inp->inp_socket, &xi.xi_socket);
723 			xi.xi_inp.inp_gencnt = inp->inp_gencnt;
724 			INP_RUNLOCK(inp);
725 			error = SYSCTL_OUT(req, &xi, sizeof xi);
726 		} else
727 			INP_RUNLOCK(inp);
728 	}
729 	if (!error) {
730 		/*
731 		 * Give the user an updated idea of our state.  If the
732 		 * generation differs from what we told her before, she knows
733 		 * that something happened while we were processing this
734 		 * request, and it might be necessary to retry.
735 		 */
736 		INP_INFO_RLOCK(&V_udbinfo);
737 		xig.xig_gen = V_udbinfo.ipi_gencnt;
738 		xig.xig_sogen = so_gencnt;
739 		xig.xig_count = V_udbinfo.ipi_count;
740 		INP_INFO_RUNLOCK(&V_udbinfo);
741 		error = SYSCTL_OUT(req, &xig, sizeof xig);
742 	}
743 	free(inp_list, M_TEMP);
744 	return (error);
745 }
746 
747 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
748     udp_pcblist, "S,xinpcb", "List of active UDP sockets");
749 
750 static int
751 udp_getcred(SYSCTL_HANDLER_ARGS)
752 {
753 	INIT_VNET_INET(curvnet);
754 	struct xucred xuc;
755 	struct sockaddr_in addrs[2];
756 	struct inpcb *inp;
757 	int error;
758 
759 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
760 	if (error)
761 		return (error);
762 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
763 	if (error)
764 		return (error);
765 	INP_INFO_RLOCK(&V_udbinfo);
766 	inp = in_pcblookup_hash(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port,
767 				addrs[0].sin_addr, addrs[0].sin_port, 1, NULL);
768 	if (inp != NULL) {
769 		INP_RLOCK(inp);
770 		INP_INFO_RUNLOCK(&V_udbinfo);
771 		if (inp->inp_socket == NULL)
772 			error = ENOENT;
773 		if (error == 0)
774 			error = cr_canseeinpcb(req->td->td_ucred, inp);
775 		if (error == 0)
776 			cru2x(inp->inp_cred, &xuc);
777 		INP_RUNLOCK(inp);
778 	} else {
779 		INP_INFO_RUNLOCK(&V_udbinfo);
780 		error = ENOENT;
781 	}
782 	if (error == 0)
783 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
784 	return (error);
785 }
786 
787 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred,
788     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
789     udp_getcred, "S,xucred", "Get the xucred of a UDP connection");
790 
791 static int
792 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
793     struct mbuf *control, struct thread *td)
794 {
795 	INIT_VNET_INET(inp->inp_vnet);
796 	struct udpiphdr *ui;
797 	int len = m->m_pkthdr.len;
798 	struct in_addr faddr, laddr;
799 	struct cmsghdr *cm;
800 	struct sockaddr_in *sin, src;
801 	int error = 0;
802 	int ipflags;
803 	u_short fport, lport;
804 	int unlock_udbinfo;
805 
806 	/*
807 	 * udp_output() may need to temporarily bind or connect the current
808 	 * inpcb.  As such, we don't know up front whether we will need the
809 	 * pcbinfo lock or not.  Do any work to decide what is needed up
810 	 * front before acquiring any locks.
811 	 */
812 	if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
813 		if (control)
814 			m_freem(control);
815 		m_freem(m);
816 		return (EMSGSIZE);
817 	}
818 
819 	src.sin_family = 0;
820 	if (control != NULL) {
821 		/*
822 		 * XXX: Currently, we assume all the optional information is
823 		 * stored in a single mbuf.
824 		 */
825 		if (control->m_next) {
826 			m_freem(control);
827 			m_freem(m);
828 			return (EINVAL);
829 		}
830 		for (; control->m_len > 0;
831 		    control->m_data += CMSG_ALIGN(cm->cmsg_len),
832 		    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
833 			cm = mtod(control, struct cmsghdr *);
834 			if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
835 			    || cm->cmsg_len > control->m_len) {
836 				error = EINVAL;
837 				break;
838 			}
839 			if (cm->cmsg_level != IPPROTO_IP)
840 				continue;
841 
842 			switch (cm->cmsg_type) {
843 			case IP_SENDSRCADDR:
844 				if (cm->cmsg_len !=
845 				    CMSG_LEN(sizeof(struct in_addr))) {
846 					error = EINVAL;
847 					break;
848 				}
849 				bzero(&src, sizeof(src));
850 				src.sin_family = AF_INET;
851 				src.sin_len = sizeof(src);
852 				src.sin_port = inp->inp_lport;
853 				src.sin_addr =
854 				    *(struct in_addr *)CMSG_DATA(cm);
855 				break;
856 
857 			default:
858 				error = ENOPROTOOPT;
859 				break;
860 			}
861 			if (error)
862 				break;
863 		}
864 		m_freem(control);
865 	}
866 	if (error) {
867 		m_freem(m);
868 		return (error);
869 	}
870 
871 	/*
872 	 * Depending on whether or not the application has bound or connected
873 	 * the socket, we may have to do varying levels of work.  The optimal
874 	 * case is for a connected UDP socket, as a global lock isn't
875 	 * required at all.
876 	 *
877 	 * In order to decide which we need, we require stability of the
878 	 * inpcb binding, which we ensure by acquiring a read lock on the
879 	 * inpcb.  This doesn't strictly follow the lock order, so we play
880 	 * the trylock and retry game; note that we may end up with more
881 	 * conservative locks than required the second time around, so later
882 	 * assertions have to accept that.  Further analysis of the number of
883 	 * misses under contention is required.
884 	 */
885 	sin = (struct sockaddr_in *)addr;
886 	INP_RLOCK(inp);
887 	if (sin != NULL &&
888 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
889 		INP_RUNLOCK(inp);
890 		INP_INFO_WLOCK(&V_udbinfo);
891 		INP_WLOCK(inp);
892 		unlock_udbinfo = 2;
893 	} else if ((sin != NULL && (
894 	    (sin->sin_addr.s_addr == INADDR_ANY) ||
895 	    (sin->sin_addr.s_addr == INADDR_BROADCAST) ||
896 	    (inp->inp_laddr.s_addr == INADDR_ANY) ||
897 	    (inp->inp_lport == 0))) ||
898 	    (src.sin_family == AF_INET)) {
899 		if (!INP_INFO_TRY_RLOCK(&V_udbinfo)) {
900 			INP_RUNLOCK(inp);
901 			INP_INFO_RLOCK(&V_udbinfo);
902 			INP_RLOCK(inp);
903 		}
904 		unlock_udbinfo = 1;
905 	} else
906 		unlock_udbinfo = 0;
907 
908 	/*
909 	 * If the IP_SENDSRCADDR control message was specified, override the
910 	 * source address for this datagram.  Its use is invalidated if the
911 	 * address thus specified is incomplete or clobbers other inpcbs.
912 	 */
913 	laddr = inp->inp_laddr;
914 	lport = inp->inp_lport;
915 	if (src.sin_family == AF_INET) {
916 		INP_INFO_LOCK_ASSERT(&V_udbinfo);
917 		if ((lport == 0) ||
918 		    (laddr.s_addr == INADDR_ANY &&
919 		     src.sin_addr.s_addr == INADDR_ANY)) {
920 			error = EINVAL;
921 			goto release;
922 		}
923 		error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
924 		    &laddr.s_addr, &lport, td->td_ucred);
925 		if (error)
926 			goto release;
927 	}
928 
929 	/*
930 	 * If a UDP socket has been connected, then a local address/port will
931 	 * have been selected and bound.
932 	 *
933 	 * If a UDP socket has not been connected to, then an explicit
934 	 * destination address must be used, in which case a local
935 	 * address/port may not have been selected and bound.
936 	 */
937 	if (sin != NULL) {
938 		INP_LOCK_ASSERT(inp);
939 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
940 			error = EISCONN;
941 			goto release;
942 		}
943 
944 		/*
945 		 * Jail may rewrite the destination address, so let it do
946 		 * that before we use it.
947 		 */
948 		error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
949 		if (error)
950 			goto release;
951 
952 		/*
953 		 * If a local address or port hasn't yet been selected, or if
954 		 * the destination address needs to be rewritten due to using
955 		 * a special INADDR_ constant, invoke in_pcbconnect_setup()
956 		 * to do the heavy lifting.  Once a port is selected, we
957 		 * commit the binding back to the socket; we also commit the
958 		 * binding of the address if in jail.
959 		 *
960 		 * If we already have a valid binding and we're not
961 		 * requesting a destination address rewrite, use a fast path.
962 		 */
963 		if (inp->inp_laddr.s_addr == INADDR_ANY ||
964 		    inp->inp_lport == 0 ||
965 		    sin->sin_addr.s_addr == INADDR_ANY ||
966 		    sin->sin_addr.s_addr == INADDR_BROADCAST) {
967 			INP_INFO_LOCK_ASSERT(&V_udbinfo);
968 			error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
969 			    &lport, &faddr.s_addr, &fport, NULL,
970 			    td->td_ucred);
971 			if (error)
972 				goto release;
973 
974 			/*
975 			 * XXXRW: Why not commit the port if the address is
976 			 * !INADDR_ANY?
977 			 */
978 			/* Commit the local port if newly assigned. */
979 			if (inp->inp_laddr.s_addr == INADDR_ANY &&
980 			    inp->inp_lport == 0) {
981 				INP_INFO_WLOCK_ASSERT(&V_udbinfo);
982 				INP_WLOCK_ASSERT(inp);
983 				/*
984 				 * Remember addr if jailed, to prevent
985 				 * rebinding.
986 				 */
987 				if (jailed(td->td_ucred))
988 					inp->inp_laddr = laddr;
989 				inp->inp_lport = lport;
990 				if (in_pcbinshash(inp) != 0) {
991 					inp->inp_lport = 0;
992 					error = EAGAIN;
993 					goto release;
994 				}
995 				inp->inp_flags |= INP_ANONPORT;
996 			}
997 		} else {
998 			faddr = sin->sin_addr;
999 			fport = sin->sin_port;
1000 		}
1001 	} else {
1002 		INP_LOCK_ASSERT(inp);
1003 		faddr = inp->inp_faddr;
1004 		fport = inp->inp_fport;
1005 		if (faddr.s_addr == INADDR_ANY) {
1006 			error = ENOTCONN;
1007 			goto release;
1008 		}
1009 	}
1010 
1011 	/*
1012 	 * Calculate data length and get a mbuf for UDP, IP, and possible
1013 	 * link-layer headers.  Immediate slide the data pointer back forward
1014 	 * since we won't use that space at this layer.
1015 	 */
1016 	M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_DONTWAIT);
1017 	if (m == NULL) {
1018 		error = ENOBUFS;
1019 		goto release;
1020 	}
1021 	m->m_data += max_linkhdr;
1022 	m->m_len -= max_linkhdr;
1023 	m->m_pkthdr.len -= max_linkhdr;
1024 
1025 	/*
1026 	 * Fill in mbuf with extended UDP header and addresses and length put
1027 	 * into network format.
1028 	 */
1029 	ui = mtod(m, struct udpiphdr *);
1030 	bzero(ui->ui_x1, sizeof(ui->ui_x1));	/* XXX still needed? */
1031 	ui->ui_pr = IPPROTO_UDP;
1032 	ui->ui_src = laddr;
1033 	ui->ui_dst = faddr;
1034 	ui->ui_sport = lport;
1035 	ui->ui_dport = fport;
1036 	ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
1037 
1038 	/*
1039 	 * Set the Don't Fragment bit in the IP header.
1040 	 */
1041 	if (inp->inp_flags & INP_DONTFRAG) {
1042 		struct ip *ip;
1043 
1044 		ip = (struct ip *)&ui->ui_i;
1045 		ip->ip_off |= IP_DF;
1046 	}
1047 
1048 	ipflags = 0;
1049 	if (inp->inp_socket->so_options & SO_DONTROUTE)
1050 		ipflags |= IP_ROUTETOIF;
1051 	if (inp->inp_socket->so_options & SO_BROADCAST)
1052 		ipflags |= IP_ALLOWBROADCAST;
1053 	if (inp->inp_flags & INP_ONESBCAST)
1054 		ipflags |= IP_SENDONES;
1055 
1056 #ifdef MAC
1057 	mac_inpcb_create_mbuf(inp, m);
1058 #endif
1059 
1060 	/*
1061 	 * Set up checksum and output datagram.
1062 	 */
1063 	if (udp_cksum) {
1064 		if (inp->inp_flags & INP_ONESBCAST)
1065 			faddr.s_addr = INADDR_BROADCAST;
1066 		ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
1067 		    htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP));
1068 		m->m_pkthdr.csum_flags = CSUM_UDP;
1069 		m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1070 	} else
1071 		ui->ui_sum = 0;
1072 	((struct ip *)ui)->ip_len = sizeof (struct udpiphdr) + len;
1073 	((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl;	/* XXX */
1074 	((struct ip *)ui)->ip_tos = inp->inp_ip_tos;	/* XXX */
1075 	UDPSTAT_INC(udps_opackets);
1076 
1077 	if (unlock_udbinfo == 2)
1078 		INP_INFO_WUNLOCK(&V_udbinfo);
1079 	else if (unlock_udbinfo == 1)
1080 		INP_INFO_RUNLOCK(&V_udbinfo);
1081 	error = ip_output(m, inp->inp_options, NULL, ipflags,
1082 	    inp->inp_moptions, inp);
1083 	if (unlock_udbinfo == 2)
1084 		INP_WUNLOCK(inp);
1085 	else
1086 		INP_RUNLOCK(inp);
1087 	return (error);
1088 
1089 release:
1090 	if (unlock_udbinfo == 2) {
1091 		INP_WUNLOCK(inp);
1092 		INP_INFO_WUNLOCK(&V_udbinfo);
1093 	} else if (unlock_udbinfo == 1) {
1094 		INP_RUNLOCK(inp);
1095 		INP_INFO_RUNLOCK(&V_udbinfo);
1096 	} else
1097 		INP_RUNLOCK(inp);
1098 	m_freem(m);
1099 	return (error);
1100 }
1101 
1102 static void
1103 udp_abort(struct socket *so)
1104 {
1105 	INIT_VNET_INET(so->so_vnet);
1106 	struct inpcb *inp;
1107 
1108 	inp = sotoinpcb(so);
1109 	KASSERT(inp != NULL, ("udp_abort: inp == NULL"));
1110 	INP_INFO_WLOCK(&V_udbinfo);
1111 	INP_WLOCK(inp);
1112 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1113 		in_pcbdisconnect(inp);
1114 		inp->inp_laddr.s_addr = INADDR_ANY;
1115 		soisdisconnected(so);
1116 	}
1117 	INP_WUNLOCK(inp);
1118 	INP_INFO_WUNLOCK(&V_udbinfo);
1119 }
1120 
1121 static int
1122 udp_attach(struct socket *so, int proto, struct thread *td)
1123 {
1124 	INIT_VNET_INET(so->so_vnet);
1125 	struct inpcb *inp;
1126 	int error;
1127 
1128 	inp = sotoinpcb(so);
1129 	KASSERT(inp == NULL, ("udp_attach: inp != NULL"));
1130 	error = soreserve(so, udp_sendspace, udp_recvspace);
1131 	if (error)
1132 		return (error);
1133 	INP_INFO_WLOCK(&V_udbinfo);
1134 	error = in_pcballoc(so, &V_udbinfo);
1135 	if (error) {
1136 		INP_INFO_WUNLOCK(&V_udbinfo);
1137 		return (error);
1138 	}
1139 
1140 	inp = (struct inpcb *)so->so_pcb;
1141 	INP_INFO_WUNLOCK(&V_udbinfo);
1142 	inp->inp_vflag |= INP_IPV4;
1143 	inp->inp_ip_ttl = V_ip_defttl;
1144 	/*
1145 	 * UDP does not have a per-protocol pcb (inp->inp_ppcb).
1146 	 * We use this pointer for kernel tunneling pointer.
1147 	 * If we ever need to have a protocol block we will
1148 	 * need to move this function pointer there. Null
1149 	 * in this pointer means "do the normal thing".
1150 	 */
1151 	inp->inp_ppcb = NULL;
1152 	INP_WUNLOCK(inp);
1153 	return (0);
1154 }
1155 
1156 int
1157 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f)
1158 {
1159 	struct inpcb *inp;
1160 
1161 	inp = (struct inpcb *)so->so_pcb;
1162 	KASSERT(so->so_type == SOCK_DGRAM, ("udp_set_kernel_tunneling: !dgram"));
1163 	KASSERT(so->so_pcb != NULL, ("udp_set_kernel_tunneling: NULL inp"));
1164 	if (so->so_type != SOCK_DGRAM) {
1165 		/* Not UDP socket... sorry! */
1166 		return (ENOTSUP);
1167 	}
1168 	if (inp == NULL) {
1169 		/* NULL INP? */
1170 		return (EINVAL);
1171 	}
1172 	INP_WLOCK(inp);
1173 	if (inp->inp_ppcb != NULL) {
1174 		INP_WUNLOCK(inp);
1175 		return (EBUSY);
1176 	}
1177 	inp->inp_ppcb = f;
1178 	INP_WUNLOCK(inp);
1179 	return (0);
1180 }
1181 
1182 static int
1183 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
1184 {
1185 	INIT_VNET_INET(so->so_vnet);
1186 	struct inpcb *inp;
1187 	int error;
1188 
1189 	inp = sotoinpcb(so);
1190 	KASSERT(inp != NULL, ("udp_bind: inp == NULL"));
1191 	INP_INFO_WLOCK(&V_udbinfo);
1192 	INP_WLOCK(inp);
1193 	error = in_pcbbind(inp, nam, td->td_ucred);
1194 	INP_WUNLOCK(inp);
1195 	INP_INFO_WUNLOCK(&V_udbinfo);
1196 	return (error);
1197 }
1198 
1199 static void
1200 udp_close(struct socket *so)
1201 {
1202 	INIT_VNET_INET(so->so_vnet);
1203 	struct inpcb *inp;
1204 
1205 	inp = sotoinpcb(so);
1206 	KASSERT(inp != NULL, ("udp_close: inp == NULL"));
1207 	INP_INFO_WLOCK(&V_udbinfo);
1208 	INP_WLOCK(inp);
1209 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1210 		in_pcbdisconnect(inp);
1211 		inp->inp_laddr.s_addr = INADDR_ANY;
1212 		soisdisconnected(so);
1213 	}
1214 	INP_WUNLOCK(inp);
1215 	INP_INFO_WUNLOCK(&V_udbinfo);
1216 }
1217 
1218 static int
1219 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1220 {
1221 	INIT_VNET_INET(so->so_vnet);
1222 	struct inpcb *inp;
1223 	int error;
1224 	struct sockaddr_in *sin;
1225 
1226 	inp = sotoinpcb(so);
1227 	KASSERT(inp != NULL, ("udp_connect: inp == NULL"));
1228 	INP_INFO_WLOCK(&V_udbinfo);
1229 	INP_WLOCK(inp);
1230 	if (inp->inp_faddr.s_addr != INADDR_ANY) {
1231 		INP_WUNLOCK(inp);
1232 		INP_INFO_WUNLOCK(&V_udbinfo);
1233 		return (EISCONN);
1234 	}
1235 	sin = (struct sockaddr_in *)nam;
1236 	error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
1237 	if (error != 0) {
1238 		INP_WUNLOCK(inp);
1239 		INP_INFO_WUNLOCK(&V_udbinfo);
1240 		return (error);
1241 	}
1242 	error = in_pcbconnect(inp, nam, td->td_ucred);
1243 	if (error == 0)
1244 		soisconnected(so);
1245 	INP_WUNLOCK(inp);
1246 	INP_INFO_WUNLOCK(&V_udbinfo);
1247 	return (error);
1248 }
1249 
1250 static void
1251 udp_detach(struct socket *so)
1252 {
1253 	INIT_VNET_INET(so->so_vnet);
1254 	struct inpcb *inp;
1255 
1256 	inp = sotoinpcb(so);
1257 	KASSERT(inp != NULL, ("udp_detach: inp == NULL"));
1258 	KASSERT(inp->inp_faddr.s_addr == INADDR_ANY,
1259 	    ("udp_detach: not disconnected"));
1260 	INP_INFO_WLOCK(&V_udbinfo);
1261 	INP_WLOCK(inp);
1262 	in_pcbdetach(inp);
1263 	in_pcbfree(inp);
1264 	INP_INFO_WUNLOCK(&V_udbinfo);
1265 }
1266 
1267 static int
1268 udp_disconnect(struct socket *so)
1269 {
1270 	INIT_VNET_INET(so->so_vnet);
1271 	struct inpcb *inp;
1272 
1273 	inp = sotoinpcb(so);
1274 	KASSERT(inp != NULL, ("udp_disconnect: inp == NULL"));
1275 	INP_INFO_WLOCK(&V_udbinfo);
1276 	INP_WLOCK(inp);
1277 	if (inp->inp_faddr.s_addr == INADDR_ANY) {
1278 		INP_WUNLOCK(inp);
1279 		INP_INFO_WUNLOCK(&V_udbinfo);
1280 		return (ENOTCONN);
1281 	}
1282 
1283 	in_pcbdisconnect(inp);
1284 	inp->inp_laddr.s_addr = INADDR_ANY;
1285 	SOCK_LOCK(so);
1286 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
1287 	SOCK_UNLOCK(so);
1288 	INP_WUNLOCK(inp);
1289 	INP_INFO_WUNLOCK(&V_udbinfo);
1290 	return (0);
1291 }
1292 
1293 static int
1294 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
1295     struct mbuf *control, struct thread *td)
1296 {
1297 	struct inpcb *inp;
1298 
1299 	inp = sotoinpcb(so);
1300 	KASSERT(inp != NULL, ("udp_send: inp == NULL"));
1301 	return (udp_output(inp, m, addr, control, td));
1302 }
1303 
1304 int
1305 udp_shutdown(struct socket *so)
1306 {
1307 	struct inpcb *inp;
1308 
1309 	inp = sotoinpcb(so);
1310 	KASSERT(inp != NULL, ("udp_shutdown: inp == NULL"));
1311 	INP_WLOCK(inp);
1312 	socantsendmore(so);
1313 	INP_WUNLOCK(inp);
1314 	return (0);
1315 }
1316 
1317 struct pr_usrreqs udp_usrreqs = {
1318 	.pru_abort =		udp_abort,
1319 	.pru_attach =		udp_attach,
1320 	.pru_bind =		udp_bind,
1321 	.pru_connect =		udp_connect,
1322 	.pru_control =		in_control,
1323 	.pru_detach =		udp_detach,
1324 	.pru_disconnect =	udp_disconnect,
1325 	.pru_peeraddr =		in_getpeeraddr,
1326 	.pru_send =		udp_send,
1327 	.pru_soreceive =	soreceive_dgram,
1328 	.pru_sosend =		sosend_dgram,
1329 	.pru_shutdown =		udp_shutdown,
1330 	.pru_sockaddr =		in_getsockaddr,
1331 	.pru_sosetlabel =	in_pcbsosetlabel,
1332 	.pru_close =		udp_close,
1333 };
1334